Pipeline run

{
  "final_skills": [
    {
      "is_primary": true,
      "skill_name": "MLOps"
    },
    {
      "is_primary": true,
      "skill_name": "Machine Learning"
    },
    {
      "is_primary": true,
      "skill_name": "Distributed Training"
    },
    {
      "is_primary": true,
      "skill_name": "Training Pipelines"
    },
    {
      "is_primary": true,
      "skill_name": "GPU Kernel Optimization"
    },
    {
      "is_primary": true,
      "skill_name": "ML Infrastructure"
    },
    {
      "is_primary": true,
      "skill_name": "ML Systems"
    },
    {
      "is_primary": true,
      "skill_name": "JAX"
    },
    {
      "is_primary": true,
      "skill_name": "PyTorch"
    },
    {
      "is_primary": false,
      "skill_name": "Pallas"
    },
    {
      "is_primary": false,
      "skill_name": "Triton"
    },
    {
      "is_primary": false,
      "skill_name": "ML Framework Internals"
    },
    {
      "is_primary": false,
      "skill_name": "Distributed Systems"
    }
  ],
  "jd_role": {
    "display_name": "MLOps Engineer",
    "rationale": null,
    "role_archetype": "Engineering",
    "slug": ""
  },
  "nano_parsed": {
    "JD_type": "pass",
    "about_company": {
      "source_marker": {
        "first_5_words": "Para AI Labs is the",
        "last_5_words": "including Mercor, Scale AI,"
      },
      "text": "Para AI Labs is the curated destination for high-quality AI training and evaluation roles from leading platforms including Mercor, Scale AI, Turing, and others.",
      "word_count": 27
    },
    "certifications": [],
    "company_name": "Mercor",
    "ctc": {
      "currency": "USD",
      "max": 45,
      "min": 35,
      "period": "hourly",
      "raw": "$35\u2013$45/hour USD"
    },
    "domain": {
      "primary": {
        "aliases": [],
        "domain": "IT Services \u0026 Consulting"
      },
      "secondary": null
    },
    "education": [],
    "experience": {
      "max": null,
      "min": 2,
      "raw": "2+ years of professional experience"
    },
    "job_locations": [
      {
        "aliases": [],
        "city": null,
        "country": "India",
        "state": null,
        "work_mode": "remote"
      }
    ],
    "role": "MLOps Engineer",
    "role_archetype": "Engineering",
    "roles_and_responsibilities": [
      {
        "bullet_count": 6,
        "heading": "What you\u0027ll do",
        "heading_was_present": true,
        "source_marker": {
          "first_5_words": "Design challenging tasks across MLOps,",
          "last_5_words": "gaps in ML framework internals"
        },
        "text": "Design challenging tasks across MLOps, ML infrastructure, and ML systems for use in model training\nWrite accurate, well-structured reference solutions covering distributed training, training pipelines, and GPU kernel optimization\nEvaluate submissions from other engineers and provide clear, written technical feedback\nDevelop rubrics and guidelines for assessing training pipeline design, distributed systems reasoning, and kernel-level optimization\nCollaborate with other ML subject matter experts to maintain consistency and accuracy across the training data\nGuide research and engineering teams toward closing specific knowledge gaps in ML framework internals",
        "word_count": 66
      },
      {
        "bullet_count": 6,
        "heading": "What you need",
        "heading_was_present": true,
        "source_marker": {
          "first_5_words": "2+ years of professional experience",
          "last_5_words": "on weekdays"
        },
        "text": "2+ years of professional experience in ML Infrastructure, MLOps, or ML Systems Engineering at a recognized, top-tier organization\nHands-on production experience with JAX and/or Py Torch at scale \u2014 real training workloads, not coursework or hobby projects\nExperience writing or optimizing custom GPU kernels with Pallas (JAX) or Triton\nDemonstrable career progression\nStrong written English \u2014 you can explain complex technical decisions clearly\nReliable availability for at least 30 hours/week on weekdays",
        "word_count": 66
      }
    ],
    "urls": []
  },
  "rejected": false,
  "rejection_reason": null,
  "run_id": "2f8955ed-9a4c-4f79-8590-412144b7d135",
  "stage3_signals": {
    "alias_match_roles": [
      {
        "display_name": "ML Engineer",
        "matched_count": null,
        "role_id": 3,
        "score": 0.6875,
        "slug": "ml-engineer",
        "total_count": null
      },
      {
        "display_name": "DevOps Engineer",
        "matched_count": null,
        "role_id": 10,
        "score": 0.55,
        "slug": "devops-engineer",
        "total_count": null
      },
      {
        "display_name": "Frontend Engineer",
        "matched_count": null,
        "role_id": 7,
        "score": 0.5,
        "slug": "frontend-engineer",
        "total_count": null
      },
      {
        "display_name": "AI Engineer",
        "matched_count": null,
        "role_id": 13,
        "score": 0.5,
        "slug": "ai-engineer",
        "total_count": null
      },
      {
        "display_name": "AR/VR Engineer",
        "matched_count": null,
        "role_id": 8,
        "score": 0.5,
        "slug": "ar-vr-engineer",
        "total_count": null
      }
    ],
    "kra_match_roles": [
      {
        "display_name": "ML Engineer",
        "matched_count": null,
        "role_id": 3,
        "score": 0.4771,
        "slug": "ml-engineer",
        "total_count": null
      },
      {
        "display_name": "DevOps Engineer",
        "matched_count": null,
        "role_id": 10,
        "score": 0.3801,
        "slug": "devops-engineer",
        "total_count": null
      },
      {
        "display_name": "Android Engineer",
        "matched_count": null,
        "role_id": 4,
        "score": 0.3657,
        "slug": "android-engineer",
        "total_count": null
      },
      {
        "display_name": "Backend Engineer",
        "matched_count": null,
        "role_id": 1,
        "score": 0.3565,
        "slug": "backend-engineer",
        "total_count": null
      },
      {
        "display_name": "AR/VR Engineer",
        "matched_count": null,
        "role_id": 8,
        "score": 0.3485,
        "slug": "ar-vr-engineer",
        "total_count": null
      }
    ],
    "skill_match_roles": [
      {
        "display_name": "ML Engineer",
        "matched_count": 4,
        "role_id": 3,
        "score": 0.3077,
        "slug": "ml-engineer",
        "total_count": 13
      },
      {
        "display_name": "AI Engineer",
        "matched_count": 2,
        "role_id": 13,
        "score": 0.1538,
        "slug": "ai-engineer",
        "total_count": 13
      },
      {
        "display_name": "Data Engineer",
        "matched_count": 1,
        "role_id": 2,
        "score": 0.0769,
        "slug": "data-engineer",
        "total_count": 13
      }
    ],
    "stage35_ran": false
  },
  "stage4_decision": {
    "alias_collision_detected": false,
    "case": "A",
    "chosen_role": {
      "display_name": "ML Engineer",
      "matched_count": null,
      "role_id": 3,
      "score": 0.4771,
      "slug": "ml-engineer",
      "total_count": null
    },
    "confidence": 0.4771,
    "llm2_fired": false,
    "llm2_reasoning": null,
    "queued": false,
    "reasoning": "All 3 signals top-rank ml-engineer"
  },
  "stage5_updates": {
    "centroid_n_after": 10,
    "centroid_updated": true,
    "collision_log_id": null,
    "new_kra_attached": null,
    "new_skills_attached": [
      {
        "is_primary": true,
        "queue_id": 1099,
        "role_display_name": "ML Engineer",
        "role_slug": "ml-engineer",
        "skill_name": "Distributed Training",
        "status": "pending"
      },
      {
        "is_primary": true,
        "queue_id": 1100,
        "role_display_name": "ML Engineer",
        "role_slug": "ml-engineer",
        "skill_name": "Training Pipelines",
        "status": "pending"
      },
      {
        "is_primary": true,
        "queue_id": 1101,
        "role_display_name": "ML Engineer",
        "role_slug": "ml-engineer",
        "skill_name": "GPU Kernel Optimization",
        "status": "pending"
      },
      {
        "is_primary": true,
        "queue_id": 1102,
        "role_display_name": "ML Engineer",
        "role_slug": "ml-engineer",
        "skill_name": "ML Infrastructure",
        "status": "pending"
      },
      {
        "is_primary": true,
        "queue_id": 1103,
        "role_display_name": "ML Engineer",
        "role_slug": "ml-engineer",
        "skill_name": "ML Systems",
        "status": "pending"
      },
      {
        "is_primary": false,
        "queue_id": 1104,
        "role_display_name": "ML Engineer",
        "role_slug": "ml-engineer",
        "skill_name": "Pallas",
        "status": "pending"
      },
      {
        "is_primary": false,
        "queue_id": 1105,
        "role_display_name": "ML Engineer",
        "role_slug": "ml-engineer",
        "skill_name": "Triton",
        "status": "pending"
      },
      {
        "is_primary": false,
        "queue_id": 1106,
        "role_display_name": "ML Engineer",
        "role_slug": "ml-engineer",
        "skill_name": "ML Framework Internals",
        "status": "pending"
      },
      {
        "is_primary": false,
        "queue_id": 1107,
        "role_display_name": "ML Engineer",
        "role_slug": "ml-engineer",
        "skill_name": "Distributed Systems",
        "status": "pending"
      }
    ],
    "queue_entry_id": null,
    "v3_pipeline_triggered": false,
    "v3_role_slug": null,
    "v3_run_id": null
  }
}

{
  "alias_matches": [
    {
      "alias_persist_skipped_reason": "alias_text already exists for this canonical skill",
      "alias_persisted": false,
      "existing_alias_id": 1832,
      "existing_alias_text": "MLOps",
      "input_term": "MLOps",
      "matched_canonical": {
        "category_id": 8,
        "display_name": "MLOps",
        "id": 1196,
        "is_also_category": false,
        "is_extractable": true,
        "skill_nature": "METHODOLOGY",
        "slug": "mlops",
        "sub_category_id": 906,
        "typical_lifespan": "EVERGREEN",
        "volatility": "STABLE"
      },
      "matched_via": "alias"
    },
    {
      "alias_persist_skipped_reason": "alias_text already exists for this canonical skill",
      "alias_persisted": false,
      "existing_alias_id": 2015,
      "existing_alias_text": "Machine Learning",
      "input_term": "Machine Learning",
      "matched_canonical": {
        "category_id": 2,
        "display_name": "Machine Learning",
        "id": 1356,
        "is_also_category": false,
        "is_extractable": true,
        "skill_nature": "CONCEPT",
        "slug": "machine-learning",
        "sub_category_id": 1024,
        "typical_lifespan": "EVERGREEN",
        "volatility": "STABLE"
      },
      "matched_via": "alias"
    },
    {
      "alias_persist_skipped_reason": "alias_text already exists for this canonical skill",
      "alias_persisted": false,
      "existing_alias_id": 451,
      "existing_alias_text": "JAX",
      "input_term": "JAX",
      "matched_canonical": {
        "category_id": 7,
        "display_name": "JAX",
        "id": 199,
        "is_also_category": false,
        "is_extractable": true,
        "skill_nature": "LIBRARY",
        "slug": "jax",
        "sub_category_id": 156,
        "typical_lifespan": "EVERGREEN",
        "volatility": "EMERGING"
      },
      "matched_via": "alias"
    },
    {
      "alias_persist_skipped_reason": "alias_text already exists for this canonical skill",
      "alias_persisted": false,
      "existing_alias_id": 441,
      "existing_alias_text": "PyTorch",
      "input_term": "PyTorch",
      "matched_canonical": {
        "category_id": 7,
        "display_name": "PyTorch",
        "id": 195,
        "is_also_category": false,
        "is_extractable": true,
        "skill_nature": "LIBRARY",
        "slug": "pytorch",
        "sub_category_id": 156,
        "typical_lifespan": "EVERGREEN",
        "volatility": "STABLE"
      },
      "matched_via": "alias"
    }
  ],
  "candidate_roles": [
    {
      "display_name": "ML Engineer",
      "id": 3,
      "rationale": null,
      "role_archetype": null,
      "slug": "ml-engineer",
      "source": "db"
    },
    {
      "display_name": "Data Engineer",
      "id": 2,
      "rationale": null,
      "role_archetype": null,
      "slug": "data-engineer",
      "source": "db"
    },
    {
      "display_name": "AI Engineer",
      "id": 13,
      "rationale": null,
      "role_archetype": null,
      "slug": "ai-engineer",
      "source": "db"
    },
    {
      "display_name": "DevOps Engineer",
      "id": 10,
      "rationale": null,
      "role_archetype": null,
      "slug": "devops-engineer",
      "source": "db"
    },
    {
      "display_name": "Backend Engineer",
      "id": 1,
      "rationale": null,
      "role_archetype": "A Backend Engineer designs, builds, and maintains the server-side logic and data handling that power applications and services. They focus on implementing reliable business functionality, integrating with other systems, and ensuring the backend is scalable, maintainable, and observable.",
      "slug": "backend-engineer",
      "source": "db"
    },
    {
      "display_name": "Cybersecurity Engineer",
      "id": 5,
      "rationale": null,
      "role_archetype": null,
      "slug": "cybersecurity-engineer",
      "source": "db"
    }
  ],
  "chosen_role": {
    "display_name": "ML Engineer",
    "id": 3,
    "rationale": "The primary skills revolve around MLOps and machine learning infrastructure, which aligns well with the responsibilities of an ML Engineer.",
    "role_archetype": null,
    "slug": "ml-engineer",
    "source": "db"
  },
  "dimensions": [
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "CI/CD for Machine Learning",
        "id": 56,
        "rationale": "Tools and platforms for automating ML model integration, testing, and deployment pipelines.",
        "slug": "ci-cd-for-machine-learning",
        "source": "db"
      },
      "input_skill": "MLOps",
      "llm_role": null,
      "roles_from_db": [
        {
          "display_name": "ML Engineer",
          "id": 3,
          "rationale": null,
          "role_archetype": null,
          "slug": "ml-engineer",
          "source": "db"
        }
      ]
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "Data Lineage and Metadata",
        "id": 28,
        "rationale": "Cataloging, documenting, and tracing how data moves and changes across systems. This dimension supports impact analysis, governance, discoverability, and operational understanding of datasets.",
        "slug": "data-lineage-and-metadata",
        "source": "db"
      },
      "input_skill": "MLOps",
      "llm_role": null,
      "roles_from_db": [
        {
          "display_name": "Data Engineer",
          "id": 2,
          "rationale": null,
          "role_archetype": null,
          "slug": "data-engineer",
          "source": "db"
        }
      ]
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "Deployment Rollouts and Release Control",
        "id": 51,
        "rationale": "Practices for safely promoting models through environments and managing rollback when production behavior changes. This dimension covers release gating, version pinning, and rollout strategies specific to ML systems.",
        "slug": "deployment-rollouts-and-release-control",
        "source": "db"
      },
      "input_skill": "MLOps",
      "llm_role": null,
      "roles_from_db": [
        {
          "display_name": "ML Engineer",
          "id": 3,
          "rationale": null,
          "role_archetype": null,
          "slug": "ml-engineer",
          "source": "db"
        }
      ]
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "AI Governance and Model Security",
        "id": 50,
        "rationale": "Controls and documentation used to make models safer, auditable, and compliant. ML engineers use this to manage model risk, supply chain integrity, and governance requirements.",
        "slug": "ai-governance-and-model-security",
        "source": "db"
      },
      "input_skill": "Machine Learning",
      "llm_role": null,
      "roles_from_db": [
        {
          "display_name": "AI Engineer",
          "id": 13,
          "rationale": null,
          "role_archetype": null,
          "slug": "ai-engineer",
          "source": "db"
        },
        {
          "display_name": "ML Engineer",
          "id": 3,
          "rationale": null,
          "role_archetype": null,
          "slug": "ml-engineer",
          "source": "db"
        }
      ]
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "React Frontend Development",
        "id": 96,
        "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
        "slug": "d_init_01",
        "source": "db"
      },
      "input_skill": "Machine Learning",
      "llm_role": null,
      "roles_from_db": []
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "ML Frameworks and Libraries",
        "id": 40,
        "rationale": "Core libraries used to define models, train them, run inference, and evaluate predictive performance. These frameworks shape how ML engineers express model architectures and training loops.",
        "slug": "ml-frameworks-and-libraries",
        "source": "db"
      },
      "input_skill": "JAX",
      "llm_role": null,
      "roles_from_db": [
        {
          "display_name": "ML Engineer",
          "id": 3,
          "rationale": null,
          "role_archetype": null,
          "slug": "ml-engineer",
          "source": "db"
        }
      ]
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "ML Frameworks and Libraries",
        "id": 40,
        "rationale": "Core libraries used to define models, train them, run inference, and evaluate predictive performance. These frameworks shape how ML engineers express model architectures and training loops.",
        "slug": "ml-frameworks-and-libraries",
        "source": "db"
      },
      "input_skill": "PyTorch",
      "llm_role": null,
      "roles_from_db": [
        {
          "display_name": "ML Engineer",
          "id": 3,
          "rationale": null,
          "role_archetype": null,
          "slug": "ml-engineer",
          "source": "db"
        }
      ]
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "Model Fine-Tuning \u0026 Adaptation",
        "id": 212,
        "rationale": "Techniques and libraries for adapting pre-trained language models to specific tasks or domains.",
        "slug": "model-fine-tuning-adaptation",
        "source": "db"
      },
      "input_skill": "PyTorch",
      "llm_role": null,
      "roles_from_db": [
        {
          "display_name": "AI Engineer",
          "id": 13,
          "rationale": null,
          "role_archetype": null,
          "slug": "ai-engineer",
          "source": "db"
        }
      ]
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "React Frontend Development",
        "id": 96,
        "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
        "slug": "d_init_01",
        "source": "db"
      },
      "input_skill": "Distributed Training",
      "llm_role": null,
      "roles_from_db": []
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "CI/CD Pipeline Platforms",
        "id": 150,
        "rationale": "Systems used to define, run, and maintain automated build and deployment workflows. This cluster is coherent because the role owns delivery automation end to end, including pipeline reliability and promotion logic.",
        "slug": "ci-cd-pipeline-platforms",
        "source": "db"
      },
      "input_skill": "Training Pipelines",
      "llm_role": null,
      "roles_from_db": [
        {
          "display_name": "DevOps Engineer",
          "id": 10,
          "rationale": null,
          "role_archetype": null,
          "slug": "devops-engineer",
          "source": "db"
        }
      ]
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "CI/CD Pipeline Platforms",
        "id": 150,
        "rationale": "Systems used to define, run, and maintain automated build and deployment workflows. This cluster is coherent because the role owns delivery automation end to end, including pipeline reliability and promotion logic.",
        "slug": "ci-cd-pipeline-platforms",
        "source": "db"
      },
      "input_skill": "Training Pipelines",
      "llm_role": null,
      "roles_from_db": [
        {
          "display_name": "DevOps Engineer",
          "id": 10,
          "rationale": null,
          "role_archetype": null,
          "slug": "devops-engineer",
          "source": "db"
        }
      ]
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "Model Optimization and Acceleration",
        "id": 53,
        "rationale": "Techniques for improving inference latency, throughput, memory use, and training efficiency. ML engineers use these methods to meet production constraints without sacrificing too much quality.",
        "slug": "model-optimization-and-acceleration",
        "source": "db"
      },
      "input_skill": "GPU Kernel Optimization",
      "llm_role": null,
      "roles_from_db": [
        {
          "display_name": "ML Engineer",
          "id": 3,
          "rationale": null,
          "role_archetype": null,
          "slug": "ml-engineer",
          "source": "db"
        }
      ]
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "Model Optimization and Acceleration",
        "id": 53,
        "rationale": "Techniques for improving inference latency, throughput, memory use, and training efficiency. ML engineers use these methods to meet production constraints without sacrificing too much quality.",
        "slug": "model-optimization-and-acceleration",
        "source": "db"
      },
      "input_skill": "GPU Kernel Optimization",
      "llm_role": null,
      "roles_from_db": [
        {
          "display_name": "ML Engineer",
          "id": 3,
          "rationale": null,
          "role_archetype": null,
          "slug": "ml-engineer",
          "source": "db"
        }
      ]
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "React Frontend Development",
        "id": 96,
        "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
        "slug": "d_init_01",
        "source": "db"
      },
      "input_skill": "ML Infrastructure",
      "llm_role": null,
      "roles_from_db": []
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "React Frontend Development",
        "id": 96,
        "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
        "slug": "d_init_01",
        "source": "db"
      },
      "input_skill": "ML Systems",
      "llm_role": null,
      "roles_from_db": []
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "Model Optimization and Acceleration",
        "id": 53,
        "rationale": "Techniques for improving inference latency, throughput, memory use, and training efficiency. ML engineers use these methods to meet production constraints without sacrificing too much quality.",
        "slug": "model-optimization-and-acceleration",
        "source": "db"
      },
      "input_skill": "Pallas",
      "llm_role": null,
      "roles_from_db": [
        {
          "display_name": "ML Engineer",
          "id": 3,
          "rationale": null,
          "role_archetype": null,
          "slug": "ml-engineer",
          "source": "db"
        }
      ]
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "Model Optimization and Acceleration",
        "id": 53,
        "rationale": "Techniques for improving inference latency, throughput, memory use, and training efficiency. ML engineers use these methods to meet production constraints without sacrificing too much quality.",
        "slug": "model-optimization-and-acceleration",
        "source": "db"
      },
      "input_skill": "Triton",
      "llm_role": null,
      "roles_from_db": [
        {
          "display_name": "ML Engineer",
          "id": 3,
          "rationale": null,
          "role_archetype": null,
          "slug": "ml-engineer",
          "source": "db"
        }
      ]
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "React Frontend Development",
        "id": 96,
        "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
        "slug": "d_init_01",
        "source": "db"
      },
      "input_skill": "ML Framework Internals",
      "llm_role": null,
      "roles_from_db": []
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "React Frontend Development",
        "id": 96,
        "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
        "slug": "d_init_01",
        "source": "db"
      },
      "input_skill": "Distributed Systems",
      "llm_role": null,
      "roles_from_db": []
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "Performance and Scalability Tuning",
        "id": 11,
        "rationale": "Techniques for improving throughput, latency, and resource efficiency in backend services. Focuses on profiling, concurrency, load behavior, and bottleneck elimination.",
        "slug": "performance-and-scalability-tuning",
        "source": "db"
      },
      "input_skill": "Distributed Systems",
      "llm_role": null,
      "roles_from_db": [
        {
          "display_name": "Backend Engineer",
          "id": 1,
          "rationale": null,
          "role_archetype": "A Backend Engineer designs, builds, and maintains the server-side logic and data handling that power applications and services. They focus on implementing reliable business functionality, integrating with other systems, and ensuring the backend is scalable, maintainable, and observable.",
          "slug": "backend-engineer",
          "source": "db"
        }
      ]
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "Cloud Platforms",
        "id": 20,
        "rationale": "Proficiency in major cloud service provider platforms and their core services.",
        "slug": "cloud-platforms",
        "source": "db"
      },
      "input_skill": "Distributed Systems",
      "llm_role": null,
      "roles_from_db": [
        {
          "display_name": "Backend Engineer",
          "id": 1,
          "rationale": null,
          "role_archetype": "A Backend Engineer designs, builds, and maintains the server-side logic and data handling that power applications and services. They focus on implementing reliable business functionality, integrating with other systems, and ensuring the backend is scalable, maintainable, and observable.",
          "slug": "backend-engineer",
          "source": "db"
        },
        {
          "display_name": "Cybersecurity Engineer",
          "id": 5,
          "rationale": null,
          "role_archetype": null,
          "slug": "cybersecurity-engineer",
          "source": "db"
        },
        {
          "display_name": "Data Engineer",
          "id": 2,
          "rationale": null,
          "role_archetype": null,
          "slug": "data-engineer",
          "source": "db"
        },
        {
          "display_name": "DevOps Engineer",
          "id": 10,
          "rationale": null,
          "role_archetype": null,
          "slug": "devops-engineer",
          "source": "db"
        },
        {
          "display_name": "ML Engineer",
          "id": 3,
          "rationale": null,
          "role_archetype": null,
          "slug": "ml-engineer",
          "source": "db"
        }
      ]
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "Cloud Platforms",
        "id": 20,
        "rationale": "Proficiency in major cloud service provider platforms and their core services.",
        "slug": "cloud-platforms",
        "source": "db"
      },
      "input_skill": "Distributed Systems",
      "llm_role": null,
      "roles_from_db": [
        {
          "display_name": "Backend Engineer",
          "id": 1,
          "rationale": null,
          "role_archetype": "A Backend Engineer designs, builds, and maintains the server-side logic and data handling that power applications and services. They focus on implementing reliable business functionality, integrating with other systems, and ensuring the backend is scalable, maintainable, and observable.",
          "slug": "backend-engineer",
          "source": "db"
        },
        {
          "display_name": "Cybersecurity Engineer",
          "id": 5,
          "rationale": null,
          "role_archetype": null,
          "slug": "cybersecurity-engineer",
          "source": "db"
        },
        {
          "display_name": "Data Engineer",
          "id": 2,
          "rationale": null,
          "role_archetype": null,
          "slug": "data-engineer",
          "source": "db"
        },
        {
          "display_name": "DevOps Engineer",
          "id": 10,
          "rationale": null,
          "role_archetype": null,
          "slug": "devops-engineer",
          "source": "db"
        },
        {
          "display_name": "ML Engineer",
          "id": 3,
          "rationale": null,
          "role_archetype": null,
          "slug": "ml-engineer",
          "source": "db"
        }
      ]
    },
    {
      "dimension": {
        "difficulty_hint": "well_known",
        "display_name": "Performance and Scalability Tuning",
        "id": 11,
        "rationale": "Techniques for improving throughput, latency, and resource efficiency in backend services. Focuses on profiling, concurrency, load behavior, and bottleneck elimination.",
        "slug": "performance-and-scalability-tuning",
        "source": "db"
      },
      "input_skill": "Distributed Systems",
      "llm_role": null,
      "roles_from_db": [
        {
          "display_name": "Backend Engineer",
          "id": 1,
          "rationale": null,
          "role_archetype": "A Backend Engineer designs, builds, and maintains the server-side logic and data handling that power applications and services. They focus on implementing reliable business functionality, integrating with other systems, and ensuring the backend is scalable, maintainable, and observable.",
          "slug": "backend-engineer",
          "source": "db"
        }
      ]
    }
  ],
  "input_final_skills": [
    "MLOps",
    "Machine Learning",
    "Distributed Training",
    "Training Pipelines",
    "GPU Kernel Optimization",
    "ML Infrastructure",
    "ML Systems",
    "JAX",
    "PyTorch",
    "Pallas",
    "Triton",
    "ML Framework Internals",
    "Distributed Systems"
  ],
  "input_llm_skills": [
    "MLOps",
    "Machine Learning",
    "Distributed Training",
    "Training Pipelines",
    "GPU Kernel Optimization",
    "ML Infrastructure",
    "ML Systems",
    "JAX",
    "PyTorch",
    "Pallas",
    "Triton",
    "ML Framework Internals",
    "Distributed Systems"
  ],
  "new_aliases_persisted": 0,
  "run_id": "2f8955ed-9a4c-4f79-8590-412144b7d135",
  "skills_detail": [
    {
      "aliases_in_db": [
        {
          "alias_text": "MLOps",
          "alias_type": "CANONICAL",
          "id": 1832,
          "is_primary": false,
          "match_strategy": "CASE_INSENSITIVE"
        }
      ],
      "canonical": {
        "category_id": 8,
        "display_name": "MLOps",
        "id": 1196,
        "is_also_category": false,
        "is_extractable": true,
        "skill_nature": "METHODOLOGY",
        "slug": "mlops",
        "sub_category_id": 906,
        "typical_lifespan": "EVERGREEN",
        "volatility": "STABLE"
      },
      "dimensions": [
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "CI/CD for Machine Learning",
            "id": 56,
            "rationale": "Tools and platforms for automating ML model integration, testing, and deployment pipelines.",
            "slug": "ci-cd-for-machine-learning",
            "source": "db"
          },
          "input_skill": "MLOps",
          "llm_role": null,
          "roles_from_db": [
            {
              "display_name": "ML Engineer",
              "id": 3,
              "rationale": null,
              "role_archetype": null,
              "slug": "ml-engineer",
              "source": "db"
            }
          ]
        },
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "Data Lineage and Metadata",
            "id": 28,
            "rationale": "Cataloging, documenting, and tracing how data moves and changes across systems. This dimension supports impact analysis, governance, discoverability, and operational understanding of datasets.",
            "slug": "data-lineage-and-metadata",
            "source": "db"
          },
          "input_skill": "MLOps",
          "llm_role": null,
          "roles_from_db": [
            {
              "display_name": "Data Engineer",
              "id": 2,
              "rationale": null,
              "role_archetype": null,
              "slug": "data-engineer",
              "source": "db"
            }
          ]
        },
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "Deployment Rollouts and Release Control",
            "id": 51,
            "rationale": "Practices for safely promoting models through environments and managing rollback when production behavior changes. This dimension covers release gating, version pinning, and rollout strategies specific to ML systems.",
            "slug": "deployment-rollouts-and-release-control",
            "source": "db"
          },
          "input_skill": "MLOps",
          "llm_role": null,
          "roles_from_db": [
            {
              "display_name": "ML Engineer",
              "id": 3,
              "rationale": null,
              "role_archetype": null,
              "slug": "ml-engineer",
              "source": "db"
            }
          ]
        }
      ],
      "input_skill": "MLOps",
      "matched_via": "alias",
      "new_alias_persisted": false,
      "new_alias_text": null,
      "new_skill_meta": null,
      "source_tag": "db",
      "was_in_llm_skills": true
    },
    {
      "aliases_in_db": [
        {
          "alias_text": "Machine Learning",
          "alias_type": "CANONICAL",
          "id": 2015,
          "is_primary": false,
          "match_strategy": "CASE_INSENSITIVE"
        }
      ],
      "canonical": {
        "category_id": 2,
        "display_name": "Machine Learning",
        "id": 1356,
        "is_also_category": false,
        "is_extractable": true,
        "skill_nature": "CONCEPT",
        "slug": "machine-learning",
        "sub_category_id": 1024,
        "typical_lifespan": "EVERGREEN",
        "volatility": "STABLE"
      },
      "dimensions": [
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "AI Governance and Model Security",
            "id": 50,
            "rationale": "Controls and documentation used to make models safer, auditable, and compliant. ML engineers use this to manage model risk, supply chain integrity, and governance requirements.",
            "slug": "ai-governance-and-model-security",
            "source": "db"
          },
          "input_skill": "Machine Learning",
          "llm_role": null,
          "roles_from_db": [
            {
              "display_name": "AI Engineer",
              "id": 13,
              "rationale": null,
              "role_archetype": null,
              "slug": "ai-engineer",
              "source": "db"
            },
            {
              "display_name": "ML Engineer",
              "id": 3,
              "rationale": null,
              "role_archetype": null,
              "slug": "ml-engineer",
              "source": "db"
            }
          ]
        },
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "React Frontend Development",
            "id": 96,
            "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
            "slug": "d_init_01",
            "source": "db"
          },
          "input_skill": "Machine Learning",
          "llm_role": null,
          "roles_from_db": []
        }
      ],
      "input_skill": "Machine Learning",
      "matched_via": "alias",
      "new_alias_persisted": false,
      "new_alias_text": null,
      "new_skill_meta": null,
      "source_tag": "db",
      "was_in_llm_skills": true
    },
    {
      "aliases_in_db": [],
      "canonical": null,
      "dimensions": [
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "React Frontend Development",
            "id": 96,
            "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
            "slug": "d_init_01",
            "source": "db"
          },
          "input_skill": "Distributed Training",
          "llm_role": null,
          "roles_from_db": []
        }
      ],
      "input_skill": "Distributed Training",
      "matched_via": null,
      "new_alias_persisted": false,
      "new_alias_text": null,
      "new_skill_meta": {
        "derived": {
          "category": "Architecture",
          "skill_nature": "PATTERN",
          "sub_category": "distributed_training",
          "typical_lifespan": "EVERGREEN",
          "version_strategy": "NOT_APPLICABLE",
          "volatility": "STABLE"
        },
        "enrichment": {
          "ambiguity": {
            "ambiguity_flag": false,
            "confused_with": [],
            "reasoning": "\u201cDistributed Training\u201d is a specific ML training paradigm; it\u2019s unlikely to be confused with other catalog skills."
          },
          "context_keywords": {
            "context_keywords": [
              "TensorFlow",
              "PyTorch",
              "Horovod",
              "data parallelism",
              "model parallelism",
              "gradient accumulation",
              "multi-GPU",
              "cloud computing",
              "Kubernetes",
              "scalability",
              "synchronization",
              "fault tolerance",
              "asynchronous training",
              "distributed systems",
              "high-performance computing"
            ]
          },
          "maturity": {
            "confidence": 0.86,
            "maturity": "well_known",
            "reasoning": "Common in ML/AI job descriptions for large-model work; major vendors like AWS, Google Cloud, and NVIDIA document distributed training as a standard production pattern."
          },
          "skill_id": "distributed-training",
          "vendor_license": {
            "confidence": 0.9,
            "license": null,
            "vendor": null,
            "year_introduced": null
          },
          "versioning": {
            "current_version": null,
            "version_aliases": {},
            "versioned": false
          }
        },
        "keep_log": [],
        "locked_dimensions": [
          {
            "description": "Training machine learning models across multiple GPUs, nodes, or clusters to reduce wall-clock time and scale beyond a single device. This skill belongs here because it covers the coordination, communication, and parallelism needed for distributed training runs.",
            "exemplar_skills": [
              "Distributed Training",
              "PyTorch Distributed",
              "Horovod",
              "DeepSpeed",
              "FSDP",
              "multi-GPU training",
              "multi-node training"
            ],
            "in_scope": "Distributed Training, data parallel training, model parallel training, tensor parallelism, pipeline parallelism, multi-GPU training, multi-node training, gradient synchronization, all-reduce, parameter servers, Horovod, PyTorch Distributed, DeepSpeed, FSDP, NCCL",
            "name": "Distributed Model Training",
            "out_of_scope": "Single-device model fitting, inference optimization, experiment tracking, model evaluation metrics, cloud provisioning, container orchestration, which belong to other dimensions",
            "overlap_flags": [
              {
                "reason": "Distributed training often overlaps with efficiency work such as mixed precision and memory reduction, but that dimension is broader and includes inference optimization.",
                "with_dim_id": "model-optimization-and-acceleration",
                "with_dim_name": null,
                "with_role": "ML Engineer"
              },
              {
                "reason": "Both involve parallel execution concepts, but this dimension is specific to ML training coordination rather than general programming concurrency.",
                "with_dim_id": "concurrency-and-parallel-processing",
                "with_dim_name": null,
                "with_role": "Backend Engineer"
              }
            ],
            "tentative_id": "d_init_01"
          }
        ],
        "merge_log": [],
        "placed": {
          "name": "Distributed Training",
          "placement_confidence": 0.92,
          "primary_dimension": "d_init_01",
          "reasoning": "Deterministic JD placement: locked_dimensions has 1 dimension(s) from skill-driven dimension generation after reconciliation; primary_dimension is the first locked dim.",
          "secondary_dimensions": [],
          "skill_id": "distributed-training"
        },
        "relationships": {
          "child_skills": [],
          "parent_skills": [],
          "related_to": [
            "machine-learning",
            "mlops",
            "kubernetes",
            "autoscaling",
            "load-balancing",
            "monitoring",
            "ci-cd",
            "event-driven-architecture"
          ],
          "requires": [],
          "skill_id": "distributed-training",
          "suppress_on_match": []
        },
        "skill_id": "distributed-training",
        "split_log": [],
        "typed": {
          "alternatives_considered": [],
          "confidence": 0.9,
          "name": "Distributed Training",
          "reasoning": "Distributed Training is fundamentally a system-shape pattern for how model training is organized across multiple compute nodes, so it fits the Architecture category rather than a tool or methodology.",
          "skill_id": "distributed-training",
          "subtype": "distributed_training",
          "type": "Architecture"
        },
        "warnings": []
      },
      "source_tag": "llm",
      "was_in_llm_skills": true
    },
    {
      "aliases_in_db": [],
      "canonical": null,
      "dimensions": [
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "CI/CD Pipeline Platforms",
            "id": 150,
            "rationale": "Systems used to define, run, and maintain automated build and deployment workflows. This cluster is coherent because the role owns delivery automation end to end, including pipeline reliability and promotion logic.",
            "slug": "ci-cd-pipeline-platforms",
            "source": "db"
          },
          "input_skill": "Training Pipelines",
          "llm_role": null,
          "roles_from_db": [
            {
              "display_name": "DevOps Engineer",
              "id": 10,
              "rationale": null,
              "role_archetype": null,
              "slug": "devops-engineer",
              "source": "db"
            }
          ]
        },
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "CI/CD Pipeline Platforms",
            "id": 150,
            "rationale": "Systems used to define, run, and maintain automated build and deployment workflows. This cluster is coherent because the role owns delivery automation end to end, including pipeline reliability and promotion logic.",
            "slug": "ci-cd-pipeline-platforms",
            "source": "db"
          },
          "input_skill": "Training Pipelines",
          "llm_role": null,
          "roles_from_db": [
            {
              "display_name": "DevOps Engineer",
              "id": 10,
              "rationale": null,
              "role_archetype": null,
              "slug": "devops-engineer",
              "source": "db"
            }
          ]
        }
      ],
      "input_skill": "Training Pipelines",
      "matched_via": null,
      "new_alias_persisted": false,
      "new_alias_text": null,
      "new_skill_meta": {
        "derived": {
          "category": "Architecture",
          "skill_nature": "PATTERN",
          "sub_category": "ml_training_pipeline_architecture",
          "typical_lifespan": "EVERGREEN",
          "version_strategy": "NOT_APPLICABLE",
          "volatility": "STABLE"
        },
        "enrichment": {
          "ambiguity": {
            "ambiguity_flag": true,
            "confused_with": [
              "ci-cd-pipelines",
              "mlops-deployment-pipelines"
            ],
            "reasoning": "\u201cTraining Pipelines\u201d in JDs can be used broadly for CI/CD pipeline architecture or MLOps deployment pipelines, not strictly ML training pipeline design."
          },
          "context_keywords": {
            "context_keywords": [
              "MLflow",
              "Kubeflow",
              "TensorFlow",
              "PyTorch",
              "CI/CD",
              "Docker",
              "Kubernetes",
              "data preprocessing",
              "hyperparameter tuning",
              "model deployment",
              "orchestration",
              "data versioning",
              "pipeline automation",
              "experiment tracking",
              "continuous integration"
            ]
          },
          "maturity": {
            "confidence": 0.9,
            "maturity": "well_known",
            "reasoning": "Common in ML job postings and platform docs; many JDs for MLOps/ML Engineer explicitly require building training pipelines in Airflow/Kubeflow/Vertex AI/SageMaker."
          },
          "skill_id": "training-pipelines",
          "vendor_license": {
            "confidence": 0.8,
            "license": null,
            "vendor": null,
            "year_introduced": null
          },
          "versioning": {
            "current_version": null,
            "version_aliases": {},
            "versioned": false
          }
        },
        "keep_log": [
          {
            "a_dim_id": "ci-cd-pipeline-platforms",
            "a_name": "CI/CD Pipeline Platforms",
            "a_role": "__skill_focal__",
            "b_dim_id": "ci-cd-pipeline-platforms",
            "b_name": "CI/CD Pipeline Platforms",
            "b_role": "DevOps Engineer",
            "pair_kind": "cross_role",
            "reasoning": "Dim A is ML/software pipeline orchestration: build/test/package/promote workflows, with exemplars like GitHub Actions, GitLab CI, Jenkins, Argo Workflows, Tekton, and explicit inclusion of training pipelines. Dim B is a DevOps Engineer delivery-automation cluster focused on end-to-end build/deployment workflow reliability and promotion logic. career-track: no, because a senior Argo/Tekton training-pipeline engineer is not naturally a senior DevOps delivery-automation engineer; the daily work and ownership differ.",
            "similarity": 0.8511009823719058
          }
        ],
        "locked_dimensions": [
          {
            "description": "Systems used to define, run, and maintain automated workflows that build, test, package, and promote software artifacts. Training pipelines fit here when they are implemented as orchestrated automation with repeatable steps, triggers, and environment promotion.",
            "exemplar_skills": [
              "Training Pipelines",
              "GitHub Actions",
              "GitLab CI",
              "Jenkins pipelines",
              "Argo Workflows",
              "Tekton pipelines"
            ],
            "in_scope": "Training Pipelines, GitHub Actions, GitLab CI, Jenkins, Argo Workflows, Tekton, build/test/package stages, scheduled runs, trigger-based automation, artifact promotion, pipeline parameters",
            "name": "CI/CD Pipeline Platforms",
            "out_of_scope": "Model evaluation metrics and experiment comparison, which belong to experiment tracking and evaluation; model release approvals and governance gates, which belong to release readiness and approvals or AI training and deployment controls",
            "overlap_flags": [
              {
                "reason": "Training pipelines for ML often include approval gates and controlled promotion, which can overlap with governance around model training and deployment.",
                "with_dim_id": "ai-training-and-deployment-controls",
                "with_dim_name": null,
                "with_role": "AI Compliance Officer"
              },
              {
                "reason": "ML training pipelines frequently invoke evaluation steps and record metrics, which can overlap with experiment tracking workflows.",
                "with_dim_id": "experiment-tracking-and-evaluation",
                "with_dim_name": null,
                "with_role": "ML Engineer"
              }
            ],
            "tentative_id": "ci-cd-pipeline-platforms"
          },
          {
            "description": "Systems used to define, run, and maintain automated build and deployment workflows. This cluster is coherent because the role owns delivery automation end to end, including pipeline reliability and promotion logic.",
            "exemplar_skills": [
              "CI/CD Pipeline Platforms"
            ],
            "in_scope": "Skills, tools, and practices that belong under CI/CD Pipeline Platforms for the target role, including items implied by the dimension rationale.",
            "name": "CI/CD Pipeline Platforms",
            "out_of_scope": "Adjacent clusters explicitly not owned by CI/CD Pipeline Platforms, including unrelated platforms, roles, and skill families per library policy.",
            "overlap_flags": [],
            "tentative_id": "ci-cd-pipeline-platforms"
          }
        ],
        "merge_log": [],
        "placed": {
          "name": "Training Pipelines",
          "placement_confidence": 0.92,
          "primary_dimension": "ci-cd-pipeline-platforms",
          "reasoning": "Deterministic JD placement: locked_dimensions has 2 dimension(s) from skill-driven dimension generation after reconciliation; primary_dimension is the first locked dim.",
          "secondary_dimensions": [],
          "skill_id": "training-pipelines"
        },
        "relationships": {
          "child_skills": [],
          "parent_skills": [],
          "related_to": [
            "mlops",
            "ci-cd",
            "devops",
            "deployment-runbooks",
            "rollback-automation",
            "rollback-procedures",
            "git",
            "azure-devops-pipelines"
          ],
          "requires": [],
          "skill_id": "training-pipelines",
          "suppress_on_match": []
        },
        "skill_id": "training-pipelines",
        "split_log": [],
        "typed": {
          "alternatives_considered": [],
          "confidence": 0.88,
          "name": "Training Pipelines",
          "reasoning": "Training Pipelines describes a system-shape for organizing model training workflows, so by the Architecture rule it is best treated as an architecture rather than a tool or methodology.",
          "skill_id": "training-pipelines",
          "subtype": "ml_training_pipeline_architecture",
          "type": "Architecture"
        },
        "warnings": [
          "stage3_post_filter_dropped_catalog_only_locked_dims:41-\u003e2"
        ]
      },
      "source_tag": "llm",
      "was_in_llm_skills": true
    },
    {
      "aliases_in_db": [],
      "canonical": null,
      "dimensions": [
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "Model Optimization and Acceleration",
            "id": 53,
            "rationale": "Techniques for improving inference latency, throughput, memory use, and training efficiency. ML engineers use these methods to meet production constraints without sacrificing too much quality.",
            "slug": "model-optimization-and-acceleration",
            "source": "db"
          },
          "input_skill": "GPU Kernel Optimization",
          "llm_role": null,
          "roles_from_db": [
            {
              "display_name": "ML Engineer",
              "id": 3,
              "rationale": null,
              "role_archetype": null,
              "slug": "ml-engineer",
              "source": "db"
            }
          ]
        },
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "Model Optimization and Acceleration",
            "id": 53,
            "rationale": "Techniques for improving inference latency, throughput, memory use, and training efficiency. ML engineers use these methods to meet production constraints without sacrificing too much quality.",
            "slug": "model-optimization-and-acceleration",
            "source": "db"
          },
          "input_skill": "GPU Kernel Optimization",
          "llm_role": null,
          "roles_from_db": [
            {
              "display_name": "ML Engineer",
              "id": 3,
              "rationale": null,
              "role_archetype": null,
              "slug": "ml-engineer",
              "source": "db"
            }
          ]
        }
      ],
      "input_skill": "GPU Kernel Optimization",
      "matched_via": null,
      "new_alias_persisted": false,
      "new_alias_text": null,
      "new_skill_meta": {
        "derived": {
          "category": "Concept",
          "skill_nature": "CONCEPT",
          "sub_category": "gpu_kernel_optimization",
          "typical_lifespan": "EVERGREEN",
          "version_strategy": "NOT_APPLICABLE",
          "volatility": "STABLE"
        },
        "enrichment": {
          "ambiguity": {
            "ambiguity_flag": false,
            "confused_with": [],
            "reasoning": "\u201cGPU Kernel Optimization\u201d is a specific performance-tuning concept; unlikely to be confused with other distinct skills in typical JDs."
          },
          "context_keywords": {
            "context_keywords": [
              "CUDA",
              "OpenCL",
              "threading",
              "memory coalescing",
              "latency reduction",
              "profiling",
              "kernel launch",
              "shared memory",
              "register allocation",
              "SIMD",
              "performance tuning",
              "GPU architecture",
              "compute capability",
              "parallel processing",
              "algorithm optimization"
            ]
          },
          "maturity": {
            "confidence": 0.86,
            "maturity": "well_known",
            "reasoning": "Widely listed in ML/HPC job descriptions and supported by major vendor docs (NVIDIA CUDA, ROCm) as a core performance skill for production inference/training."
          },
          "skill_id": "gpu-kernel-optimization",
          "vendor_license": {
            "confidence": 0.95,
            "license": null,
            "vendor": null,
            "year_introduced": null
          },
          "versioning": {
            "current_version": null,
            "version_aliases": {},
            "versioned": false
          }
        },
        "keep_log": [
          {
            "a_dim_id": "model-optimization-and-acceleration",
            "a_name": "Model Optimization and Acceleration",
            "a_role": "__skill_focal__",
            "b_dim_id": "model-optimization-and-acceleration",
            "b_name": "Model Optimization and Acceleration",
            "b_role": "ML Engineer",
            "pair_kind": "cross_role",
            "reasoning": "Cross-role false positive. Dim A is a low-level optimization specialty: its description explicitly includes GPU kernel optimization, and exemplars like CUDA kernels, TensorRT, ONNX Runtime, operator fusion, and mixed precision point to systems/performance engineering. Dim B is the ML Engineer version of the same label, framed around using optimization methods to meet production constraints; that is broader applied ML work, not the same specialist cluster. career-track: no, because a senior CUDA/TensorRT optimization engineer is not naturally a senior ML engineer across the full production-constraint cluster.",
            "similarity": 0.8094799927913185
          }
        ],
        "locked_dimensions": [
          {
            "description": "Techniques for improving inference latency, throughput, memory use, and training efficiency. GPU kernel optimization belongs here because it is a low-level acceleration method used to make model execution faster and more resource efficient.",
            "exemplar_skills": [
              "GPU Kernel Optimization",
              "CUDA kernels",
              "operator fusion",
              "mixed precision",
              "quantization",
              "TensorRT",
              "ONNX Runtime"
            ],
            "in_scope": "GPU Kernel Optimization, quantization, pruning, distillation, operator fusion, graph optimization, TensorRT, ONNX Runtime, CUDA kernels, mixed precision, memory bandwidth reduction, latency tuning",
            "name": "Model Optimization and Acceleration",
            "out_of_scope": "Cloud deployment, model governance, experiment tracking, data labeling, frontend performance, general backend scalability, which belong to other dimensions",
            "overlap_flags": [
              {
                "reason": "Both involve performance improvement, but this dimension is specific to model execution and accelerator-level optimization rather than general service tuning.",
                "with_dim_id": "performance-and-scalability-tuning",
                "with_dim_name": null,
                "with_role": "Backend Engineer"
              },
              {
                "reason": "Kernel work uses parallel execution concepts, but the focus here is GPU-specific acceleration and memory behavior rather than general concurrency patterns.",
                "with_dim_id": "concurrency-and-parallel-processing",
                "with_dim_name": null,
                "with_role": "Backend Engineer"
              }
            ],
            "tentative_id": "model-optimization-and-acceleration"
          },
          {
            "description": "Techniques for improving inference latency, throughput, memory use, and training efficiency. ML engineers use these methods to meet production constraints without sacrificing too much quality.",
            "exemplar_skills": [
              "Model Optimization and Acceleration"
            ],
            "in_scope": "Skills, tools, and practices that belong under Model Optimization and Acceleration for the target role, including items implied by the dimension rationale.",
            "name": "Model Optimization and Acceleration",
            "out_of_scope": "Adjacent clusters explicitly not owned by Model Optimization and Acceleration, including unrelated platforms, roles, and skill families per library policy.",
            "overlap_flags": [],
            "tentative_id": "model-optimization-and-acceleration"
          }
        ],
        "merge_log": [],
        "placed": {
          "name": "GPU Kernel Optimization",
          "placement_confidence": 0.92,
          "primary_dimension": "model-optimization-and-acceleration",
          "reasoning": "Deterministic JD placement: locked_dimensions has 2 dimension(s) from skill-driven dimension generation after reconciliation; primary_dimension is the first locked dim.",
          "secondary_dimensions": [],
          "skill_id": "gpu-kernel-optimization"
        },
        "relationships": {
          "child_skills": [],
          "parent_skills": [
            "pytorch"
          ],
          "related_to": [
            "mlops",
            "kubernetes",
            "load-balancing",
            "build-variants",
            "git",
            "jvm",
            "kotlin"
          ],
          "requires": [
            "startup-profiling",
            "memory-trimming",
            "algorithms"
          ],
          "skill_id": "gpu-kernel-optimization",
          "suppress_on_match": []
        },
        "skill_id": "gpu-kernel-optimization",
        "split_log": [],
        "typed": {
          "alternatives_considered": [],
          "confidence": 0.9,
          "name": "GPU Kernel Optimization",
          "reasoning": "This is best treated as a Concept because it names a technical knowledge area/practice about improving GPU kernel performance, not a specific tool, framework, or methodology.",
          "skill_id": "gpu-kernel-optimization",
          "subtype": "gpu_kernel_optimization",
          "type": "Concept"
        },
        "warnings": [
          "stage3_post_filter_dropped_catalog_only_locked_dims:41-\u003e2"
        ]
      },
      "source_tag": "llm",
      "was_in_llm_skills": true
    },
    {
      "aliases_in_db": [],
      "canonical": null,
      "dimensions": [
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "React Frontend Development",
            "id": 96,
            "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
            "slug": "d_init_01",
            "source": "db"
          },
          "input_skill": "ML Infrastructure",
          "llm_role": null,
          "roles_from_db": []
        }
      ],
      "input_skill": "ML Infrastructure",
      "matched_via": null,
      "new_alias_persisted": false,
      "new_alias_text": null,
      "new_skill_meta": {
        "derived": {
          "category": "Architecture",
          "skill_nature": "PATTERN",
          "sub_category": "ml_infrastructure_architecture",
          "typical_lifespan": "EVERGREEN",
          "version_strategy": "NOT_APPLICABLE",
          "volatility": "EMERGING"
        },
        "enrichment": {
          "ambiguity": {
            "ambiguity_flag": false,
            "confused_with": [],
            "reasoning": "\u201cML Infrastructure\u201d is a specific architecture domain; unlikely to be confused with other distinct catalog skills."
          },
          "context_keywords": {
            "context_keywords": [
              "Kubernetes",
              "Docker",
              "TensorFlow",
              "model serving",
              "data pipelines",
              "CI/CD",
              "cloud computing",
              "scalability",
              "monitoring",
              "orchestration",
              "microservices",
              "data lakes",
              "MLOps",
              "API management",
              "versioning"
            ]
          },
          "maturity": {
            "confidence": 0.84,
            "maturity": "emerging",
            "reasoning": "ML Infrastructure appears increasingly in JDs for MLOps/platform roles, but it is not yet a universal hiring staple like AWS or Kubernetes; market demand is growing alongside managed ML platforms."
          },
          "skill_id": "ml-infrastructure",
          "vendor_license": {
            "confidence": 0.8,
            "license": null,
            "vendor": null,
            "year_introduced": null
          },
          "versioning": {
            "current_version": null,
            "version_aliases": {},
            "versioned": false
          }
        },
        "keep_log": [],
        "locked_dimensions": [
          {
            "description": "Platform components and operational practices used to build, run, and support machine learning systems in production. This covers the infrastructure layer around training, deployment, serving, monitoring, and lifecycle automation, which is broader than model development itself.",
            "exemplar_skills": [
              "ML Infrastructure",
              "model serving",
              "feature stores",
              "model registry",
              "training orchestration",
              "inference infrastructure",
              "ML deployment pipelines",
              "model monitoring"
            ],
            "in_scope": "ML Infrastructure, model serving platforms, training orchestration, feature stores, model registries, inference endpoints, GPU/accelerator provisioning, workflow automation, deployment pipelines for ML, monitoring for model and data drift",
            "name": "ML Infrastructure",
            "out_of_scope": "Model architecture design and algorithm selection, experiment analysis and metric comparison, prompt engineering and LLM prompt versioning, cloud account governance and policy controls, generic CI/CD not specific to ML workloads",
            "overlap_flags": [
              {
                "reason": "ML infrastructure often uses CI/CD systems for packaging and deployment, but the ML-specific runtime, registry, and serving concerns are distinct.",
                "with_dim_id": "ci-cd-pipeline-platforms",
                "with_dim_name": null,
                "with_role": "DevOps Engineer"
              },
              {
                "reason": "ML infrastructure is commonly implemented on cloud services, but this dimension focuses on ML workload patterns rather than general cloud proficiency.",
                "with_dim_id": "cloud-platforms",
                "with_dim_name": null,
                "with_role": "Backend Engineer, Cybersecurity Engineer, Data Engineer, DevOps Engineer, ML Engineer"
              },
              {
                "reason": "Production ML infrastructure may include latency and throughput tuning, but optimization techniques themselves belong to the model performance dimension.",
                "with_dim_id": "model-optimization-and-acceleration",
                "with_dim_name": null,
                "with_role": "ML Engineer"
              }
            ],
            "tentative_id": "d_init_01"
          }
        ],
        "merge_log": [],
        "placed": {
          "name": "ML Infrastructure",
          "placement_confidence": 0.92,
          "primary_dimension": "d_init_01",
          "reasoning": "Deterministic JD placement: locked_dimensions has 1 dimension(s) from skill-driven dimension generation after reconciliation; primary_dimension is the first locked dim.",
          "secondary_dimensions": [],
          "skill_id": "ml-infrastructure"
        },
        "relationships": {
          "child_skills": [],
          "parent_skills": [],
          "related_to": [
            "mlops",
            "machine-learning",
            "ci-cd",
            "apis",
            "istio",
            "mtls",
            "triton-inference-server",
            "aws-bedrock"
          ],
          "requires": [],
          "skill_id": "ml-infrastructure",
          "suppress_on_match": []
        },
        "skill_id": "ml-infrastructure",
        "split_log": [],
        "typed": {
          "alternatives_considered": [
            "Domain: ruled out \u2014 it is broader than a vertical or industry body of knowledge and instead describes how systems are organized.",
            "Concept: ruled out \u2014 the term is an implementation-oriented system shape, not just a knowledge unit."
          ],
          "confidence": 0.78,
          "name": "ML Infrastructure",
          "reasoning": "By the Architecture vs Concept rule, \"ML Infrastructure\" names a system-shape / supporting architecture for building and operating ML systems rather than a single tool, platform, or concept.",
          "skill_id": "ml-infrastructure",
          "subtype": "ml_infrastructure_architecture",
          "type": "Architecture"
        },
        "warnings": []
      },
      "source_tag": "llm",
      "was_in_llm_skills": true
    },
    {
      "aliases_in_db": [],
      "canonical": null,
      "dimensions": [
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "React Frontend Development",
            "id": 96,
            "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
            "slug": "d_init_01",
            "source": "db"
          },
          "input_skill": "ML Systems",
          "llm_role": null,
          "roles_from_db": []
        }
      ],
      "input_skill": "ML Systems",
      "matched_via": null,
      "new_alias_persisted": false,
      "new_alias_text": null,
      "new_skill_meta": {
        "derived": {
          "category": "Domain",
          "skill_nature": "CONCEPT",
          "sub_category": "machine_learning_systems",
          "typical_lifespan": "EVERGREEN",
          "version_strategy": "NOT_APPLICABLE",
          "volatility": "EMERGING"
        },
        "enrichment": {
          "ambiguity": {
            "ambiguity_flag": false,
            "confused_with": [],
            "reasoning": "\u201cML Systems\u201d is a specific domain phrase referring to machine learning systems; it\u2019s unlikely to be confused with another distinct catalog skill."
          },
          "context_keywords": {
            "context_keywords": [
              "scalability",
              "model serving",
              "data pipelines",
              "feature engineering",
              "A/B testing",
              "monitoring",
              "hyperparameter tuning",
              "distributed training",
              "MLOps",
              "cloud deployment",
              "containerization",
              "TensorFlow",
              "PyTorch",
              "Kubeflow",
              "data versioning"
            ]
          },
          "maturity": {
            "confidence": 0.84,
            "maturity": "emerging",
            "reasoning": "ML Systems appears increasingly in JDs for MLOps/platform roles, but it is still far less universal than core cloud or backend skills; market demand is growing with production ML tooling adoption."
          },
          "skill_id": "ml-systems",
          "vendor_license": {
            "confidence": 0.9,
            "license": null,
            "vendor": null,
            "year_introduced": null
          },
          "versioning": {
            "current_version": null,
            "version_aliases": {},
            "versioned": false
          }
        },
        "keep_log": [],
        "locked_dimensions": [
          {
            "description": "End-to-end systems concerns for building, serving, and operating ML workloads in production. This includes the infrastructure, runtime, and workflow patterns that make training and inference reliable, scalable, and maintainable, which is why ML Systems fits here.",
            "exemplar_skills": [
              "ML Systems",
              "model serving",
              "feature pipelines",
              "training infrastructure",
              "inference infrastructure",
              "distributed training",
              "model deployment",
              "GPU scheduling"
            ],
            "in_scope": "ML Systems, model serving architectures, feature pipelines, training/inference infrastructure, batch and online inference, model packaging, deployment workflows, distributed training orchestration, GPU/accelerator utilization, model lifecycle automation",
            "name": "Machine Learning Systems",
            "out_of_scope": "Prompt engineering and prompt templates, model governance and compliance controls, experiment-only analysis without production systems, generic cloud provisioning not specific to ML workloads",
            "overlap_flags": [
              {
                "reason": "Production ML systems often include latency, throughput, and accelerator-efficiency work, but that dimension is specifically about optimization techniques rather than the broader system architecture.",
                "with_dim_id": "model-optimization-and-acceleration",
                "with_dim_name": null,
                "with_role": "ML Engineer"
              },
              {
                "reason": "ML systems commonly integrate experiment tracking and evaluation, but that dimension centers on measurement and comparison rather than deployment and runtime concerns.",
                "with_dim_id": "experiment-tracking-and-evaluation",
                "with_dim_name": null,
                "with_role": "ML Engineer"
              },
              {
                "reason": "ML systems may use CI/CD for model delivery, but the pipeline-platform dimension is broader and not ML-specific.",
                "with_dim_id": "ci-cd-pipeline-platforms",
                "with_dim_name": null,
                "with_role": "DevOps Engineer"
              }
            ],
            "tentative_id": "d_init_01"
          }
        ],
        "merge_log": [],
        "placed": {
          "name": "ML Systems",
          "placement_confidence": 0.92,
          "primary_dimension": "d_init_01",
          "reasoning": "Deterministic JD placement: locked_dimensions has 1 dimension(s) from skill-driven dimension generation after reconciliation; primary_dimension is the first locked dim.",
          "secondary_dimensions": [],
          "skill_id": "ml-systems"
        },
        "relationships": {
          "child_skills": [],
          "parent_skills": [
            "machine-learning",
            "mlops"
          ],
          "related_to": [
            "llms",
            "qlora",
            "feature-modules",
            "intelligent-automation",
            "langsmith",
            "mtls"
          ],
          "requires": [
            "algorithms",
            "data-structures",
            "rdbms"
          ],
          "skill_id": "ml-systems",
          "suppress_on_match": []
        },
        "skill_id": "ml-systems",
        "split_log": [],
        "typed": {
          "alternatives_considered": [],
          "confidence": 0.9,
          "name": "ML Systems",
          "reasoning": "This is best treated as a Domain because it names a problem-space/body of knowledge about building and operating machine learning systems, not a specific architecture, tool, or methodology.",
          "skill_id": "ml-systems",
          "subtype": "machine_learning_systems",
          "type": "Domain"
        },
        "warnings": []
      },
      "source_tag": "llm",
      "was_in_llm_skills": true
    },
    {
      "aliases_in_db": [
        {
          "alias_text": "JAX",
          "alias_type": "CANONICAL",
          "id": 451,
          "is_primary": true,
          "match_strategy": "CASE_INSENSITIVE"
        }
      ],
      "canonical": {
        "category_id": 7,
        "display_name": "JAX",
        "id": 199,
        "is_also_category": false,
        "is_extractable": true,
        "skill_nature": "LIBRARY",
        "slug": "jax",
        "sub_category_id": 156,
        "typical_lifespan": "EVERGREEN",
        "volatility": "EMERGING"
      },
      "dimensions": [
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "ML Frameworks and Libraries",
            "id": 40,
            "rationale": "Core libraries used to define models, train them, run inference, and evaluate predictive performance. These frameworks shape how ML engineers express model architectures and training loops.",
            "slug": "ml-frameworks-and-libraries",
            "source": "db"
          },
          "input_skill": "JAX",
          "llm_role": null,
          "roles_from_db": [
            {
              "display_name": "ML Engineer",
              "id": 3,
              "rationale": null,
              "role_archetype": null,
              "slug": "ml-engineer",
              "source": "db"
            }
          ]
        }
      ],
      "input_skill": "JAX",
      "matched_via": "alias",
      "new_alias_persisted": false,
      "new_alias_text": null,
      "new_skill_meta": null,
      "source_tag": "db",
      "was_in_llm_skills": true
    },
    {
      "aliases_in_db": [
        {
          "alias_text": "PyTorch",
          "alias_type": "CANONICAL",
          "id": 441,
          "is_primary": true,
          "match_strategy": "CASE_INSENSITIVE"
        }
      ],
      "canonical": {
        "category_id": 7,
        "display_name": "PyTorch",
        "id": 195,
        "is_also_category": false,
        "is_extractable": true,
        "skill_nature": "LIBRARY",
        "slug": "pytorch",
        "sub_category_id": 156,
        "typical_lifespan": "EVERGREEN",
        "volatility": "STABLE"
      },
      "dimensions": [
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "ML Frameworks and Libraries",
            "id": 40,
            "rationale": "Core libraries used to define models, train them, run inference, and evaluate predictive performance. These frameworks shape how ML engineers express model architectures and training loops.",
            "slug": "ml-frameworks-and-libraries",
            "source": "db"
          },
          "input_skill": "PyTorch",
          "llm_role": null,
          "roles_from_db": [
            {
              "display_name": "ML Engineer",
              "id": 3,
              "rationale": null,
              "role_archetype": null,
              "slug": "ml-engineer",
              "source": "db"
            }
          ]
        },
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "Model Fine-Tuning \u0026 Adaptation",
            "id": 212,
            "rationale": "Techniques and libraries for adapting pre-trained language models to specific tasks or domains.",
            "slug": "model-fine-tuning-adaptation",
            "source": "db"
          },
          "input_skill": "PyTorch",
          "llm_role": null,
          "roles_from_db": [
            {
              "display_name": "AI Engineer",
              "id": 13,
              "rationale": null,
              "role_archetype": null,
              "slug": "ai-engineer",
              "source": "db"
            }
          ]
        }
      ],
      "input_skill": "PyTorch",
      "matched_via": "alias",
      "new_alias_persisted": false,
      "new_alias_text": null,
      "new_skill_meta": null,
      "source_tag": "db",
      "was_in_llm_skills": true
    },
    {
      "aliases_in_db": [],
      "canonical": null,
      "dimensions": [
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "Model Optimization and Acceleration",
            "id": 53,
            "rationale": "Techniques for improving inference latency, throughput, memory use, and training efficiency. ML engineers use these methods to meet production constraints without sacrificing too much quality.",
            "slug": "model-optimization-and-acceleration",
            "source": "db"
          },
          "input_skill": "Pallas",
          "llm_role": null,
          "roles_from_db": [
            {
              "display_name": "ML Engineer",
              "id": 3,
              "rationale": null,
              "role_archetype": null,
              "slug": "ml-engineer",
              "source": "db"
            }
          ]
        }
      ],
      "input_skill": "Pallas",
      "matched_via": null,
      "new_alias_persisted": false,
      "new_alias_text": null,
      "new_skill_meta": {
        "derived": {
          "category": "Framework",
          "skill_nature": "FRAMEWORK",
          "sub_category": "machine_learning_framework",
          "typical_lifespan": "EVERGREEN",
          "version_strategy": "NOT_APPLICABLE",
          "volatility": "EMERGING"
        },
        "enrichment": {
          "ambiguity": {
            "ambiguity_flag": false,
            "confused_with": [],
            "reasoning": "\u201cPallas\u201d is a relatively specific ML framework name; typical JDs wouldn\u2019t confuse it with other catalog skills."
          },
          "context_keywords": {
            "context_keywords": [
              "model tuning",
              "hyperparameter",
              "gradient descent",
              "neural networks",
              "tensor optimization",
              "distributed training",
              "performance benchmarking",
              "data preprocessing",
              "feature engineering",
              "scalability",
              "real-time inference",
              "model compression",
              "transfer learning",
              "automated ML",
              "cloud deployment"
            ]
          },
          "maturity": {
            "confidence": 0.78,
            "maturity": "emerging",
            "reasoning": "Pallas is a newer JAX/TPU kernel framework from Google; it appears in niche ML systems discussions and docs, but JD volume is still low versus established accelerators like CUDA/XLA."
          },
          "skill_id": "pallas",
          "vendor_license": {
            "confidence": 0.85,
            "license": "apache_2",
            "vendor": "Pallas AI",
            "year_introduced": 2021
          },
          "versioning": {
            "current_version": null,
            "version_aliases": {},
            "versioned": false
          }
        },
        "keep_log": [],
        "locked_dimensions": [
          {
            "description": "Techniques and libraries used to make ML inference and training faster, smaller, or more memory efficient. Pallas belongs here because it is a JAX-oriented kernel programming tool used to write high-performance custom computations.",
            "exemplar_skills": [
              "Pallas",
              "JAX",
              "custom GPU kernels",
              "TPU optimization",
              "kernel fusion",
              "XLA optimization",
              "memory optimization",
              "inference acceleration"
            ],
            "in_scope": "Pallas, JAX kernel programming, custom GPU/TPU kernels, fused operations, memory-efficient tensor computation, low-level performance tuning for ML workloads, XLA-oriented acceleration, CUDA-like kernel design for JAX",
            "name": "Model Optimization and Acceleration",
            "out_of_scope": "Model serving platforms, batch orchestration, feature stores, experiment tracking, distributed training frameworks, which belong to deployment, data, or training-control dimensions",
            "overlap_flags": [
              {
                "reason": "Pallas involves writing code, but the distinguishing concern is performance-oriented ML kernel optimization rather than general scripting or application programming.",
                "with_dim_id": "programming-languages-and-scripting",
                "with_dim_name": null,
                "with_role": "Cybersecurity Engineer"
              },
              {
                "reason": "Kernel-level parallel execution concepts can overlap, but this dimension is specifically about ML acceleration and hardware-aware optimization.",
                "with_dim_id": "concurrency-and-parallel-processing",
                "with_dim_name": null,
                "with_role": "Backend Engineer"
              }
            ],
            "tentative_id": "model-optimization-and-acceleration"
          }
        ],
        "merge_log": [],
        "placed": {
          "name": "Pallas",
          "placement_confidence": 0.92,
          "primary_dimension": "model-optimization-and-acceleration",
          "reasoning": "Deterministic JD placement: locked_dimensions has 1 dimension(s) from skill-driven dimension generation after reconciliation; primary_dimension is the first locked dim.",
          "secondary_dimensions": [],
          "skill_id": "pallas"
        },
        "relationships": {
          "child_skills": [],
          "parent_skills": [],
          "related_to": [
            "peft",
            "openai",
            "pinecone",
            "spark",
            "cassandra",
            "sentry",
            "axios",
            "jest"
          ],
          "requires": [],
          "skill_id": "pallas",
          "suppress_on_match": []
        },
        "skill_id": "pallas",
        "split_log": [],
        "typed": {
          "alternatives_considered": [
            "Library: ruled out \u2014 Pallas is more than a simple importable utility and is typically used as a structured programming framework.",
            "Tool: ruled out \u2014 it is not software primarily operated by a user as a standalone application.",
            "Concept: ruled out \u2014 the name refers to an implementation framework rather than a knowledge unit."
          ],
          "confidence": 0.62,
          "name": "Pallas",
          "reasoning": "Pallas is best fit as a Framework because it is a software framework for building GPU-accelerated computations rather than a standalone tool or hosted platform.",
          "skill_id": "pallas",
          "subtype": "machine_learning_framework",
          "type": "Framework"
        },
        "warnings": []
      },
      "source_tag": "llm",
      "was_in_llm_skills": true
    },
    {
      "aliases_in_db": [],
      "canonical": null,
      "dimensions": [
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "Model Optimization and Acceleration",
            "id": 53,
            "rationale": "Techniques for improving inference latency, throughput, memory use, and training efficiency. ML engineers use these methods to meet production constraints without sacrificing too much quality.",
            "slug": "model-optimization-and-acceleration",
            "source": "db"
          },
          "input_skill": "Triton",
          "llm_role": null,
          "roles_from_db": [
            {
              "display_name": "ML Engineer",
              "id": 3,
              "rationale": null,
              "role_archetype": null,
              "slug": "ml-engineer",
              "source": "db"
            }
          ]
        }
      ],
      "input_skill": "Triton",
      "matched_via": null,
      "new_alias_persisted": false,
      "new_alias_text": null,
      "new_skill_meta": {
        "derived": {
          "category": "Language",
          "skill_nature": "LANGUAGE",
          "sub_category": "gpu_kernel_language",
          "typical_lifespan": "EVERGREEN",
          "version_strategy": "NOT_APPLICABLE",
          "volatility": "EMERGING"
        },
        "enrichment": {
          "ambiguity": {
            "ambiguity_flag": false,
            "confused_with": [],
            "reasoning": "\u201cTriton\u201d most commonly refers to the OpenAI GPU kernel language for model optimization; unlikely to be confused with other catalog skills."
          },
          "context_keywords": {
            "context_keywords": [
              "GPU",
              "kernel",
              "model optimization",
              "acceleration",
              "parallel computing",
              "CUDA",
              "performance tuning",
              "compute shaders",
              "deep learning",
              "tensor operations",
              "memory management",
              "synchronization",
              "profiling",
              "algorithm optimization",
              "hardware acceleration"
            ]
          },
          "maturity": {
            "confidence": 0.84,
            "maturity": "emerging",
            "reasoning": "Triton appears in growing ML systems job postings and is increasingly used in open-source GPU kernel repos, but it is still far from a universal hiring staple like CUDA."
          },
          "skill_id": "triton",
          "vendor_license": {
            "confidence": 0.85,
            "license": "other_open",
            "vendor": "NVIDIA",
            "year_introduced": 2015
          },
          "versioning": {
            "current_version": null,
            "version_aliases": {},
            "versioned": false
          }
        },
        "keep_log": [],
        "locked_dimensions": [
          {
            "description": "Techniques and tooling for making model inference faster, smaller, and more efficient on target hardware. Triton belongs here because it is commonly used to implement high-performance GPU kernels and accelerate deep learning workloads.",
            "exemplar_skills": [
              "Triton",
              "CUDA",
              "kernel fusion",
              "mixed precision",
              "GPU optimization",
              "inference acceleration"
            ],
            "in_scope": "Triton, custom GPU kernels, kernel fusion, memory coalescing, mixed precision, inference latency reduction, throughput optimization, CUDA-aware optimization",
            "name": "Model Optimization and Acceleration",
            "out_of_scope": "Model training workflows, experiment tracking, prompt engineering, vector search, cloud deployment orchestration, general Python application code",
            "overlap_flags": [
              {
                "reason": "Triton code is written in a Python-like DSL, so it can overlap with general programming-language skills even though the primary focus is GPU acceleration.",
                "with_dim_id": "programming-languages-and-scripting",
                "with_dim_name": null,
                "with_role": "Cybersecurity Engineer"
              },
              {
                "reason": "Triton kernels rely on parallel execution concepts, but this dimension is specifically about GPU-level optimization rather than general concurrency patterns.",
                "with_dim_id": "concurrency-and-parallel-processing",
                "with_dim_name": null,
                "with_role": "Backend Engineer"
              }
            ],
            "tentative_id": "model-optimization-and-acceleration"
          }
        ],
        "merge_log": [],
        "placed": {
          "name": "Triton",
          "placement_confidence": 0.92,
          "primary_dimension": "model-optimization-and-acceleration",
          "reasoning": "Deterministic JD placement: locked_dimensions has 1 dimension(s) from skill-driven dimension generation after reconciliation; primary_dimension is the first locked dim.",
          "secondary_dimensions": [],
          "skill_id": "triton"
        },
        "relationships": {
          "child_skills": [],
          "parent_skills": [],
          "related_to": [
            "triton-inference-server",
            "typescript",
            "apollo-client",
            "axios",
            "s3",
            "pinecone",
            "cilium",
            "istio"
          ],
          "requires": [],
          "skill_id": "triton",
          "suppress_on_match": []
        },
        "skill_id": "triton",
        "split_log": [],
        "typed": {
          "alternatives_considered": [],
          "confidence": 0.9,
          "name": "Triton",
          "reasoning": "Triton is fundamentally a programming language for writing GPU kernels, so it fits the Language type rather than a library or tool.",
          "skill_id": "triton",
          "subtype": "gpu_kernel_language",
          "type": "Language"
        },
        "warnings": []
      },
      "source_tag": "llm",
      "was_in_llm_skills": true
    },
    {
      "aliases_in_db": [],
      "canonical": null,
      "dimensions": [
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "React Frontend Development",
            "id": 96,
            "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
            "slug": "d_init_01",
            "source": "db"
          },
          "input_skill": "ML Framework Internals",
          "llm_role": null,
          "roles_from_db": []
        }
      ],
      "input_skill": "ML Framework Internals",
      "matched_via": null,
      "new_alias_persisted": false,
      "new_alias_text": null,
      "new_skill_meta": {
        "derived": {
          "category": "Concept",
          "skill_nature": "CONCEPT",
          "sub_category": "ml_framework_internals",
          "typical_lifespan": "EVERGREEN",
          "version_strategy": "NOT_APPLICABLE",
          "volatility": "STABLE"
        },
        "enrichment": {
          "ambiguity": {
            "ambiguity_flag": false,
            "confused_with": [],
            "reasoning": "\u201cML Framework Internals\u201d is a specific concept about how ML frameworks work internally, not commonly confused with other catalog skills."
          },
          "context_keywords": {
            "context_keywords": [
              "TensorFlow",
              "PyTorch",
              "model optimization",
              "backpropagation",
              "computational graph",
              "autograd",
              "GPU acceleration",
              "layer normalization",
              "data pipeline",
              "training loop",
              "gradient descent",
              "memory management",
              "performance tuning",
              "distributed training",
              "API design"
            ]
          },
          "maturity": {
            "confidence": 0.86,
            "maturity": "niche",
            "reasoning": "Job postings rarely ask for ML framework internals explicitly; it appears mostly in specialized roles at framework vendors or infra teams, while mainstream JDs focus on using PyTorch/TensorFlow rather than their internals."
          },
          "skill_id": "ml-framework-internals",
          "vendor_license": {
            "confidence": 1.0,
            "license": null,
            "vendor": null,
            "year_introduced": null
          },
          "versioning": {
            "current_version": null,
            "version_aliases": {},
            "versioned": false
          }
        },
        "keep_log": [],
        "locked_dimensions": [
          {
            "description": "Core implementation details of machine learning frameworks, including execution graphs, tensor ops, autograd, kernels, and runtime behavior. This fits the target skill because it focuses on how frameworks like PyTorch, TensorFlow, JAX, and similar systems work under the hood.",
            "exemplar_skills": [
              "ML Framework Internals",
              "PyTorch internals",
              "TensorFlow internals",
              "JAX internals",
              "autograd",
              "computation graphs",
              "custom operators",
              "tensor runtime"
            ],
            "in_scope": "ML Framework Internals, PyTorch internals, TensorFlow internals, JAX internals, autograd engines, computation graphs, tensor execution, operator dispatch, custom ops, runtime memory management, graph compilation, eager execution, framework extension points",
            "name": "ML Framework Internals",
            "out_of_scope": "Model training workflows and experiment management, covered by experiment-tracking-and-evaluation; inference latency optimization and quantization, covered by model-optimization-and-acceleration; distributed data processing, covered by ETL and ELT Tooling or concurrency-and-parallel-processing",
            "overlap_flags": [
              {
                "reason": "Framework internals often touch performance features, but that dimension owns latency, throughput, and deployment-oriented optimization techniques.",
                "with_dim_id": "model-optimization-and-acceleration",
                "with_dim_name": null,
                "with_role": "ML Engineer"
              },
              {
                "reason": "Framework runtimes may use parallel execution, but this dimension owns general-purpose concurrency primitives and coordination patterns.",
                "with_dim_id": "concurrency-and-parallel-processing",
                "with_dim_name": null,
                "with_role": "Backend Engineer"
              }
            ],
            "tentative_id": "d_init_01"
          }
        ],
        "merge_log": [],
        "placed": {
          "name": "ML Framework Internals",
          "placement_confidence": 0.92,
          "primary_dimension": "d_init_01",
          "reasoning": "Deterministic JD placement: locked_dimensions has 1 dimension(s) from skill-driven dimension generation after reconciliation; primary_dimension is the first locked dim.",
          "secondary_dimensions": [],
          "skill_id": "ml-framework-internals"
        },
        "relationships": {
          "child_skills": [],
          "parent_skills": [],
          "related_to": [
            "mlops",
            "model-versioning",
            "qlora",
            "feature-modules",
            "memory-leaks",
            "iam",
            "jvm",
            "java-interoperability"
          ],
          "requires": [],
          "skill_id": "ml-framework-internals",
          "suppress_on_match": []
        },
        "skill_id": "ml-framework-internals",
        "split_log": [],
        "typed": {
          "alternatives_considered": [],
          "confidence": 0.9,
          "name": "ML Framework Internals",
          "reasoning": "This is a named knowledge unit about how machine learning frameworks work internally, so by the Concept vs Methodology rule it is a Concept rather than a Framework or Tool.",
          "skill_id": "ml-framework-internals",
          "subtype": "ml_framework_internals",
          "type": "Concept"
        },
        "warnings": []
      },
      "source_tag": "llm",
      "was_in_llm_skills": true
    },
    {
      "aliases_in_db": [],
      "canonical": null,
      "dimensions": [
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "React Frontend Development",
            "id": 96,
            "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
            "slug": "d_init_01",
            "source": "db"
          },
          "input_skill": "Distributed Systems",
          "llm_role": null,
          "roles_from_db": []
        },
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "Performance and Scalability Tuning",
            "id": 11,
            "rationale": "Techniques for improving throughput, latency, and resource efficiency in backend services. Focuses on profiling, concurrency, load behavior, and bottleneck elimination.",
            "slug": "performance-and-scalability-tuning",
            "source": "db"
          },
          "input_skill": "Distributed Systems",
          "llm_role": null,
          "roles_from_db": [
            {
              "display_name": "Backend Engineer",
              "id": 1,
              "rationale": null,
              "role_archetype": "A Backend Engineer designs, builds, and maintains the server-side logic and data handling that power applications and services. They focus on implementing reliable business functionality, integrating with other systems, and ensuring the backend is scalable, maintainable, and observable.",
              "slug": "backend-engineer",
              "source": "db"
            }
          ]
        },
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "Cloud Platforms",
            "id": 20,
            "rationale": "Proficiency in major cloud service provider platforms and their core services.",
            "slug": "cloud-platforms",
            "source": "db"
          },
          "input_skill": "Distributed Systems",
          "llm_role": null,
          "roles_from_db": [
            {
              "display_name": "Backend Engineer",
              "id": 1,
              "rationale": null,
              "role_archetype": "A Backend Engineer designs, builds, and maintains the server-side logic and data handling that power applications and services. They focus on implementing reliable business functionality, integrating with other systems, and ensuring the backend is scalable, maintainable, and observable.",
              "slug": "backend-engineer",
              "source": "db"
            },
            {
              "display_name": "Cybersecurity Engineer",
              "id": 5,
              "rationale": null,
              "role_archetype": null,
              "slug": "cybersecurity-engineer",
              "source": "db"
            },
            {
              "display_name": "Data Engineer",
              "id": 2,
              "rationale": null,
              "role_archetype": null,
              "slug": "data-engineer",
              "source": "db"
            },
            {
              "display_name": "DevOps Engineer",
              "id": 10,
              "rationale": null,
              "role_archetype": null,
              "slug": "devops-engineer",
              "source": "db"
            },
            {
              "display_name": "ML Engineer",
              "id": 3,
              "rationale": null,
              "role_archetype": null,
              "slug": "ml-engineer",
              "source": "db"
            }
          ]
        },
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "Cloud Platforms",
            "id": 20,
            "rationale": "Proficiency in major cloud service provider platforms and their core services.",
            "slug": "cloud-platforms",
            "source": "db"
          },
          "input_skill": "Distributed Systems",
          "llm_role": null,
          "roles_from_db": [
            {
              "display_name": "Backend Engineer",
              "id": 1,
              "rationale": null,
              "role_archetype": "A Backend Engineer designs, builds, and maintains the server-side logic and data handling that power applications and services. They focus on implementing reliable business functionality, integrating with other systems, and ensuring the backend is scalable, maintainable, and observable.",
              "slug": "backend-engineer",
              "source": "db"
            },
            {
              "display_name": "Cybersecurity Engineer",
              "id": 5,
              "rationale": null,
              "role_archetype": null,
              "slug": "cybersecurity-engineer",
              "source": "db"
            },
            {
              "display_name": "Data Engineer",
              "id": 2,
              "rationale": null,
              "role_archetype": null,
              "slug": "data-engineer",
              "source": "db"
            },
            {
              "display_name": "DevOps Engineer",
              "id": 10,
              "rationale": null,
              "role_archetype": null,
              "slug": "devops-engineer",
              "source": "db"
            },
            {
              "display_name": "ML Engineer",
              "id": 3,
              "rationale": null,
              "role_archetype": null,
              "slug": "ml-engineer",
              "source": "db"
            }
          ]
        },
        {
          "dimension": {
            "difficulty_hint": "well_known",
            "display_name": "Performance and Scalability Tuning",
            "id": 11,
            "rationale": "Techniques for improving throughput, latency, and resource efficiency in backend services. Focuses on profiling, concurrency, load behavior, and bottleneck elimination.",
            "slug": "performance-and-scalability-tuning",
            "source": "db"
          },
          "input_skill": "Distributed Systems",
          "llm_role": null,
          "roles_from_db": [
            {
              "display_name": "Backend Engineer",
              "id": 1,
              "rationale": null,
              "role_archetype": "A Backend Engineer designs, builds, and maintains the server-side logic and data handling that power applications and services. They focus on implementing reliable business functionality, integrating with other systems, and ensuring the backend is scalable, maintainable, and observable.",
              "slug": "backend-engineer",
              "source": "db"
            }
          ]
        }
      ],
      "input_skill": "Distributed Systems",
      "matched_via": null,
      "new_alias_persisted": false,
      "new_alias_text": null,
      "new_skill_meta": {
        "derived": {
          "category": "Concept",
          "skill_nature": "CONCEPT",
          "sub_category": "distributed_systems",
          "typical_lifespan": "EVERGREEN",
          "version_strategy": "NOT_APPLICABLE",
          "volatility": "STABLE"
        },
        "enrichment": {
          "ambiguity": {
            "ambiguity_flag": false,
            "confused_with": [],
            "reasoning": "\u201cDistributed Systems\u201d is a broad, distinct concept; typical JDs won\u2019t confuse it with another specific catalog skill."
          },
          "context_keywords": {
            "context_keywords": [
              "consensus algorithms",
              "CAP theorem",
              "eventual consistency",
              "microservices",
              "message queues",
              "sharding",
              "replication",
              "load balancing",
              "fault tolerance",
              "distributed databases",
              "gRPC",
              "Zookeeper",
              "Kafka",
              "MapReduce",
              "Docker Swarm"
            ]
          },
          "maturity": {
            "confidence": 0.95,
            "maturity": "well_known",
            "reasoning": "Common hiring requirement in backend/platform JDs at large tech firms; appears across AWS, Kafka, microservices, and systems roles, with strong GitHub/Stack Overflow activity and no sunset signal."
          },
          "skill_id": "distributed-systems",
          "vendor_license": {
            "confidence": 1.0,
            "license": null,
            "vendor": null,
            "year_introduced": null
          },
          "versioning": {
            "current_version": null,
            "version_aliases": {},
            "versioned": false
          }
        },
        "keep_log": [
          {
            "a_dim_id": "d_init_01",
            "a_name": "Distributed Systems Concepts",
            "a_role": "__skill_focal__",
            "b_dim_id": "performance-and-scalability-tuning",
            "b_name": "Performance and Scalability Tuning",
            "b_role": "__skill_focal__",
            "pair_kind": "intra_role",
            "reasoning": "Dim A covers distributed-systems correctness: replication, sharding, partition tolerance, consistency models, consensus, leader election, and fault tolerance. Dim B covers performance engineering: throughput, latency, resource efficiency, profiling, concurrency, load behavior, and bottleneck elimination. Example skills differ: A\u2019s \u0027leader election\u0027 and \u0027eventual consistency\u0027 are not the same as B\u2019s \u0027profiling\u0027 and \u0027bottleneck elimination\u0027. career-track: no, because a senior distributed-systems engineer is not automatically a senior performance-tuning engineer; the work and expertise are distinct.",
            "similarity": 0.5629089619325675
          },
          {
            "a_dim_id": "d_init_01",
            "a_name": "Distributed Systems Concepts",
            "a_role": "__skill_focal__",
            "b_dim_id": "cloud-platforms",
            "b_name": "Cloud Platforms",
            "b_role": "__skill_focal__",
            "pair_kind": "intra_role",
            "reasoning": "Dim A is about distributed-systems theory and architecture: replication, sharding, consistency models, consensus, leader election, and fault tolerance. Its exemplar skills are all about reasoning over multi-node system behavior. Dim B is a broad cloud-provider platform competency: proficiency in major cloud service provider platforms and their core services. Even though cloud platforms often host distributed systems, the skill clusters are different: a senior distributed-systems engineer is not automatically a senior cloud-platform practitioner, and vice versa. career-track: no, because distributed-systems design/coordination expertise does not naturally make someone a senior practitioner in provider-specific cloud platforms and core services.",
            "similarity": 0.6887093957897851
          }
        ],
        "locked_dimensions": [
          {
            "description": "Core principles for designing and reasoning about systems that run across multiple nodes. This includes coordination, replication, partitioning, consistency, and failure handling, which are central to distributed systems work in MLOps and backend platforms.",
            "exemplar_skills": [
              "Distributed Systems",
              "replication",
              "sharding",
              "consistency models",
              "fault tolerance",
              "leader election"
            ],
            "in_scope": "Distributed Systems, replication, sharding, partition tolerance, consistency models, consensus basics, leader election, distributed coordination, fault tolerance, eventual consistency",
            "name": "Distributed Systems Concepts",
            "out_of_scope": "Concurrency and Parallel Processing for in-process threading and async execution, performance and scalability tuning for optimization tactics, cloud platforms for provider-specific managed services",
            "overlap_flags": [
              {
                "reason": "Both involve coordinating work across execution units, but this dimension is about multi-node system behavior rather than in-process concurrency.",
                "with_dim_id": "concurrency-and-parallel-processing",
                "with_dim_name": null,
                "with_role": "Backend Engineer"
              },
              {
                "reason": "Scalability concerns overlap, but that dimension focuses on optimization and bottleneck reduction rather than distributed architecture fundamentals.",
                "with_dim_id": "performance-and-scalability-tuning",
                "with_dim_name": null,
                "with_role": "Backend Engineer"
              }
            ],
            "tentative_id": "d_init_01"
          },
          {
            "description": "Techniques for improving throughput, latency, and resource efficiency in systems that must scale under load. Distributed systems skills often connect here when the focus is on bottlenecks, capacity, and load behavior rather than architecture fundamentals.",
            "exemplar_skills": [
              "Distributed Systems",
              "load balancing",
              "capacity planning",
              "bottleneck analysis",
              "horizontal scaling",
              "backpressure"
            ],
            "in_scope": "Distributed Systems, throughput optimization, latency reduction, load balancing, capacity planning, bottleneck analysis, caching strategies, backpressure, horizontal scaling",
            "name": "Performance and Scalability Tuning",
            "out_of_scope": "Distributed coordination, consensus, replication protocols, partitioning strategy, failure semantics, which belong to distributed systems architecture and consistency design",
            "overlap_flags": [
              {
                "reason": "Both can improve throughput, but this dimension is about system-level scaling and resource efficiency rather than programming-level parallelism.",
                "with_dim_id": "concurrency-and-parallel-processing",
                "with_dim_name": null,
                "with_role": "Backend Engineer"
              }
            ],
            "tentative_id": "performance-and-scalability-tuning"
          },
          {
            "description": "Major cloud provider environments and their core managed services. Distributed systems work in MLOps often runs on these platforms, but the platform knowledge itself is distinct from the distributed systems concepts.",
            "exemplar_skills": [
              "Distributed Systems",
              "AWS",
              "Azure",
              "GCP",
              "Kubernetes services",
              "cloud load balancers"
            ],
            "in_scope": "AWS, Azure, GCP, managed compute, managed networking, managed storage, Kubernetes services, cloud load balancers, cloud messaging services",
            "name": "Cloud Platforms",
            "out_of_scope": "Replication algorithms, consensus, partitioning, consistency models, and other distributed systems design principles, which are platform-agnostic",
            "overlap_flags": [
              {
                "reason": "Cloud systems are often provisioned with IaC, but this dimension is about the platform services themselves rather than declarative provisioning.",
                "with_dim_id": "infrastructure-as-code",
                "with_dim_name": null,
                "with_role": "Cloud Architect, DevOps Engineer"
              },
              {
                "reason": "Traffic routing and service-to-service controls can be part of distributed deployments, but this dimension is broader and platform-focused.",
                "with_dim_id": "service-mesh-and-traffic-management",
                "with_dim_name": null,
                "with_role": "Cloud Architect, DevOps Engineer"
              }
            ],
            "tentative_id": "cloud-platforms"
          },
          {
            "description": "Proficiency in major cloud service provider platforms and their core services.",
            "exemplar_skills": [
              "Cloud Platforms"
            ],
            "in_scope": "Skills, tools, and practices that belong under Cloud Platforms for the target role, including items implied by the dimension rationale.",
            "name": "Cloud Platforms",
            "out_of_scope": "Adjacent clusters explicitly not owned by Cloud Platforms, including unrelated platforms, roles, and skill families per library policy.",
            "overlap_flags": [],
            "tentative_id": "cloud-platforms"
          },
          {
            "description": "Techniques for improving throughput, latency, and resource efficiency in backend services. Focuses on profiling, concurrency, load behavior, and bottleneck elimination.",
            "exemplar_skills": [
              "Performance and Scalability Tuning"
            ],
            "in_scope": "Skills, tools, and practices that belong under Performance and Scalability Tuning for the target role, including items implied by the dimension rationale.",
            "name": "Performance and Scalability Tuning",
            "out_of_scope": "Adjacent clusters explicitly not owned by Performance and Scalability Tuning, including unrelated platforms, roles, and skill families per library policy.",
            "overlap_flags": [],
            "tentative_id": "performance-and-scalability-tuning"
          }
        ],
        "merge_log": [],
        "placed": {
          "name": "Distributed Systems",
          "placement_confidence": 0.92,
          "primary_dimension": "d_init_01",
          "reasoning": "Deterministic JD placement: locked_dimensions has 5 dimension(s) from skill-driven dimension generation after reconciliation; primary_dimension is the first locked dim.",
          "secondary_dimensions": [
            "performance-and-scalability-tuning",
            "cloud-platforms"
          ],
          "skill_id": "distributed-systems"
        },
        "relationships": {
          "child_skills": [],
          "parent_skills": [],
          "related_to": [
            "event-driven-architecture",
            "load-balancing",
            "kubernetes",
            "devops",
            "hadoop",
            "nosql",
            "rdbms",
            "data-lakes"
          ],
          "requires": [],
          "skill_id": "distributed-systems",
          "suppress_on_match": []
        },
        "skill_id": "distributed-systems",
        "split_log": [],
        "typed": {
          "alternatives_considered": [],
          "confidence": 0.98,
          "name": "Distributed Systems",
          "reasoning": "This is a knowledge unit about how systems behave across multiple machines, so by the Concept vs Architecture rule it is a Concept rather than a system-shape pattern.",
          "skill_id": "distributed-systems",
          "subtype": "distributed_systems",
          "type": "Concept"
        },
        "warnings": [
          "stage3_post_filter_dropped_catalog_only_locked_dims:43-\u003e5"
        ]
      },
      "source_tag": "llm",
      "was_in_llm_skills": true
    }
  ],
  "unmatched_skills": [
    "Distributed Training",
    "Training Pipelines",
    "GPU Kernel Optimization",
    "ML Infrastructure",
    "ML Systems",
    "Pallas",
    "Triton",
    "ML Framework Internals",
    "Distributed Systems"
  ]
}

{
  "chosen_role": {
    "display_name": "ML Engineer",
    "id": 3,
    "rationale": "The primary skills revolve around MLOps and machine learning infrastructure, which aligns well with the responsibilities of an ML Engineer.",
    "role_archetype": null,
    "slug": "ml-engineer",
    "source": "db"
  },
  "chosen_role_resolution": "in_db",
  "final_input_skills": [
    {
      "skill": "MLOps",
      "tag": "in_db"
    },
    {
      "skill": "Machine Learning",
      "tag": "in_db"
    },
    {
      "skill": "Distributed Training",
      "tag": "new"
    },
    {
      "skill": "Training Pipelines",
      "tag": "new"
    },
    {
      "skill": "GPU Kernel Optimization",
      "tag": "new"
    },
    {
      "skill": "ML Infrastructure",
      "tag": "new"
    },
    {
      "skill": "ML Systems",
      "tag": "new"
    },
    {
      "skill": "JAX",
      "tag": "in_db"
    },
    {
      "skill": "PyTorch",
      "tag": "in_db"
    },
    {
      "skill": "Pallas",
      "tag": "new"
    },
    {
      "skill": "Triton",
      "tag": "new"
    },
    {
      "skill": "ML Framework Internals",
      "tag": "new"
    },
    {
      "skill": "Distributed Systems",
      "tag": "new"
    }
  ],
  "persistence": {
    "items": [
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "CI/CD for Machine Learning",
          "id": 56,
          "rationale": "Tools and platforms for automating ML model integration, testing, and deployment pipelines.",
          "slug": "ci-cd-for-machine-learning",
          "source": "db"
        },
        "dimension_id": 56,
        "input_skill": "MLOps",
        "llm_role": null,
        "matched_chosen_role": true,
        "outcome_line": "Existing dimension (library) \u00b7 Role\u2194dimension saved",
        "role_dimension_saved": true,
        "roles_from_db": [
          {
            "display_name": "ML Engineer",
            "id": 3,
            "rationale": null,
            "role_archetype": null,
            "slug": "ml-engineer",
            "source": "db"
          }
        ],
        "skill_dimension_saved": true,
        "skill_id": 1196,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "Data Lineage and Metadata",
          "id": 28,
          "rationale": "Cataloging, documenting, and tracing how data moves and changes across systems. This dimension supports impact analysis, governance, discoverability, and operational understanding of datasets.",
          "slug": "data-lineage-and-metadata",
          "source": "db"
        },
        "dimension_id": 28,
        "input_skill": "MLOps",
        "llm_role": null,
        "matched_chosen_role": false,
        "outcome_line": "Existing dimension (library) \u00b7 Role\u2194dimension skipped (dimension not under chosen role)",
        "role_dimension_saved": false,
        "roles_from_db": [
          {
            "display_name": "Data Engineer",
            "id": 2,
            "rationale": null,
            "role_archetype": null,
            "slug": "data-engineer",
            "source": "db"
          }
        ],
        "skill_dimension_saved": true,
        "skill_id": 1196,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "Deployment Rollouts and Release Control",
          "id": 51,
          "rationale": "Practices for safely promoting models through environments and managing rollback when production behavior changes. This dimension covers release gating, version pinning, and rollout strategies specific to ML systems.",
          "slug": "deployment-rollouts-and-release-control",
          "source": "db"
        },
        "dimension_id": 51,
        "input_skill": "MLOps",
        "llm_role": null,
        "matched_chosen_role": true,
        "outcome_line": "Existing dimension (library) \u00b7 Role\u2194dimension saved",
        "role_dimension_saved": true,
        "roles_from_db": [
          {
            "display_name": "ML Engineer",
            "id": 3,
            "rationale": null,
            "role_archetype": null,
            "slug": "ml-engineer",
            "source": "db"
          }
        ],
        "skill_dimension_saved": true,
        "skill_id": 1196,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "AI Governance and Model Security",
          "id": 50,
          "rationale": "Controls and documentation used to make models safer, auditable, and compliant. ML engineers use this to manage model risk, supply chain integrity, and governance requirements.",
          "slug": "ai-governance-and-model-security",
          "source": "db"
        },
        "dimension_id": 50,
        "input_skill": "Machine Learning",
        "llm_role": null,
        "matched_chosen_role": true,
        "outcome_line": "Existing dimension (library) \u00b7 Role\u2194dimension saved",
        "role_dimension_saved": true,
        "roles_from_db": [
          {
            "display_name": "AI Engineer",
            "id": 13,
            "rationale": null,
            "role_archetype": null,
            "slug": "ai-engineer",
            "source": "db"
          },
          {
            "display_name": "ML Engineer",
            "id": 3,
            "rationale": null,
            "role_archetype": null,
            "slug": "ml-engineer",
            "source": "db"
          }
        ],
        "skill_dimension_saved": true,
        "skill_id": 1356,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "React Frontend Development",
          "id": 96,
          "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
          "slug": "d_init_01",
          "source": "db"
        },
        "dimension_id": 96,
        "input_skill": "Machine Learning",
        "llm_role": null,
        "matched_chosen_role": false,
        "outcome_line": "Existing dimension (library) \u00b7 Role\u2194dimension skipped (dimension not under chosen role)",
        "role_dimension_saved": false,
        "roles_from_db": [],
        "skill_dimension_saved": true,
        "skill_id": 1356,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "ML Frameworks and Libraries",
          "id": 40,
          "rationale": "Core libraries used to define models, train them, run inference, and evaluate predictive performance. These frameworks shape how ML engineers express model architectures and training loops.",
          "slug": "ml-frameworks-and-libraries",
          "source": "db"
        },
        "dimension_id": 40,
        "input_skill": "JAX",
        "llm_role": null,
        "matched_chosen_role": true,
        "outcome_line": "Existing dimension (library) \u00b7 Role\u2194dimension saved",
        "role_dimension_saved": true,
        "roles_from_db": [
          {
            "display_name": "ML Engineer",
            "id": 3,
            "rationale": null,
            "role_archetype": null,
            "slug": "ml-engineer",
            "source": "db"
          }
        ],
        "skill_dimension_saved": true,
        "skill_id": 199,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "ML Frameworks and Libraries",
          "id": 40,
          "rationale": "Core libraries used to define models, train them, run inference, and evaluate predictive performance. These frameworks shape how ML engineers express model architectures and training loops.",
          "slug": "ml-frameworks-and-libraries",
          "source": "db"
        },
        "dimension_id": 40,
        "input_skill": "PyTorch",
        "llm_role": null,
        "matched_chosen_role": true,
        "outcome_line": "Existing dimension (library) \u00b7 Role\u2194dimension saved",
        "role_dimension_saved": true,
        "roles_from_db": [
          {
            "display_name": "ML Engineer",
            "id": 3,
            "rationale": null,
            "role_archetype": null,
            "slug": "ml-engineer",
            "source": "db"
          }
        ],
        "skill_dimension_saved": true,
        "skill_id": 195,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "Model Fine-Tuning \u0026 Adaptation",
          "id": 212,
          "rationale": "Techniques and libraries for adapting pre-trained language models to specific tasks or domains.",
          "slug": "model-fine-tuning-adaptation",
          "source": "db"
        },
        "dimension_id": 212,
        "input_skill": "PyTorch",
        "llm_role": null,
        "matched_chosen_role": false,
        "outcome_line": "Existing dimension (library) \u00b7 Role\u2194dimension skipped (dimension not under chosen role)",
        "role_dimension_saved": false,
        "roles_from_db": [
          {
            "display_name": "AI Engineer",
            "id": 13,
            "rationale": null,
            "role_archetype": null,
            "slug": "ai-engineer",
            "source": "db"
          }
        ],
        "skill_dimension_saved": true,
        "skill_id": 195,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "React Frontend Development",
          "id": 96,
          "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
          "slug": "d_init_01",
          "source": "db"
        },
        "dimension_id": 96,
        "input_skill": "Distributed Training",
        "llm_role": null,
        "matched_chosen_role": false,
        "outcome_line": "New skill saved \u00b7 Existing dimension (library) \u00b7 Role\u2194dimension skipped (dimension not under chosen role)",
        "role_dimension_saved": false,
        "roles_from_db": [],
        "skill_dimension_saved": true,
        "skill_id": 1361,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "CI/CD Pipeline Platforms",
          "id": 150,
          "rationale": "Systems used to define, run, and maintain automated build and deployment workflows. This cluster is coherent because the role owns delivery automation end to end, including pipeline reliability and promotion logic.",
          "slug": "ci-cd-pipeline-platforms",
          "source": "db"
        },
        "dimension_id": 150,
        "input_skill": "Training Pipelines",
        "llm_role": null,
        "matched_chosen_role": false,
        "outcome_line": "New skill saved \u00b7 Existing dimension (library) \u00b7 Role\u2194dimension skipped (dimension not under chosen role)",
        "role_dimension_saved": false,
        "roles_from_db": [
          {
            "display_name": "DevOps Engineer",
            "id": 10,
            "rationale": null,
            "role_archetype": null,
            "slug": "devops-engineer",
            "source": "db"
          }
        ],
        "skill_dimension_saved": true,
        "skill_id": 1362,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "Model Optimization and Acceleration",
          "id": 53,
          "rationale": "Techniques for improving inference latency, throughput, memory use, and training efficiency. ML engineers use these methods to meet production constraints without sacrificing too much quality.",
          "slug": "model-optimization-and-acceleration",
          "source": "db"
        },
        "dimension_id": 53,
        "input_skill": "GPU Kernel Optimization",
        "llm_role": null,
        "matched_chosen_role": true,
        "outcome_line": "New skill saved \u00b7 Existing dimension (library) \u00b7 Role\u2194dimension saved",
        "role_dimension_saved": true,
        "roles_from_db": [
          {
            "display_name": "ML Engineer",
            "id": 3,
            "rationale": null,
            "role_archetype": null,
            "slug": "ml-engineer",
            "source": "db"
          }
        ],
        "skill_dimension_saved": true,
        "skill_id": 1363,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "React Frontend Development",
          "id": 96,
          "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
          "slug": "d_init_01",
          "source": "db"
        },
        "dimension_id": 96,
        "input_skill": "ML Infrastructure",
        "llm_role": null,
        "matched_chosen_role": false,
        "outcome_line": "New skill saved \u00b7 Existing dimension (library) \u00b7 Role\u2194dimension skipped (dimension not under chosen role)",
        "role_dimension_saved": false,
        "roles_from_db": [],
        "skill_dimension_saved": true,
        "skill_id": 1364,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "React Frontend Development",
          "id": 96,
          "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
          "slug": "d_init_01",
          "source": "db"
        },
        "dimension_id": 96,
        "input_skill": "ML Systems",
        "llm_role": null,
        "matched_chosen_role": false,
        "outcome_line": "New skill saved \u00b7 Existing dimension (library) \u00b7 Role\u2194dimension skipped (dimension not under chosen role)",
        "role_dimension_saved": false,
        "roles_from_db": [],
        "skill_dimension_saved": true,
        "skill_id": 1365,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "Model Optimization and Acceleration",
          "id": 53,
          "rationale": "Techniques for improving inference latency, throughput, memory use, and training efficiency. ML engineers use these methods to meet production constraints without sacrificing too much quality.",
          "slug": "model-optimization-and-acceleration",
          "source": "db"
        },
        "dimension_id": 53,
        "input_skill": "Pallas",
        "llm_role": null,
        "matched_chosen_role": true,
        "outcome_line": "New skill saved \u00b7 Existing dimension (library) \u00b7 Role\u2194dimension saved",
        "role_dimension_saved": true,
        "roles_from_db": [
          {
            "display_name": "ML Engineer",
            "id": 3,
            "rationale": null,
            "role_archetype": null,
            "slug": "ml-engineer",
            "source": "db"
          }
        ],
        "skill_dimension_saved": true,
        "skill_id": 1366,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "Model Optimization and Acceleration",
          "id": 53,
          "rationale": "Techniques for improving inference latency, throughput, memory use, and training efficiency. ML engineers use these methods to meet production constraints without sacrificing too much quality.",
          "slug": "model-optimization-and-acceleration",
          "source": "db"
        },
        "dimension_id": 53,
        "input_skill": "Triton",
        "llm_role": null,
        "matched_chosen_role": true,
        "outcome_line": "New skill saved \u00b7 Existing dimension (library) \u00b7 Role\u2194dimension saved",
        "role_dimension_saved": true,
        "roles_from_db": [
          {
            "display_name": "ML Engineer",
            "id": 3,
            "rationale": null,
            "role_archetype": null,
            "slug": "ml-engineer",
            "source": "db"
          }
        ],
        "skill_dimension_saved": true,
        "skill_id": 1367,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "React Frontend Development",
          "id": 96,
          "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
          "slug": "d_init_01",
          "source": "db"
        },
        "dimension_id": 96,
        "input_skill": "ML Framework Internals",
        "llm_role": null,
        "matched_chosen_role": false,
        "outcome_line": "New skill saved \u00b7 Existing dimension (library) \u00b7 Role\u2194dimension skipped (dimension not under chosen role)",
        "role_dimension_saved": false,
        "roles_from_db": [],
        "skill_dimension_saved": true,
        "skill_id": 1368,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "React Frontend Development",
          "id": 96,
          "rationale": "Building interactive web user interfaces with React.js, including component composition, state management, hooks, and rendering patterns. React.js belongs here because it is a core library for client-side UI development in modern web applications.",
          "slug": "d_init_01",
          "source": "db"
        },
        "dimension_id": 96,
        "input_skill": "Distributed Systems",
        "llm_role": null,
        "matched_chosen_role": false,
        "outcome_line": "New skill saved \u00b7 Existing dimension (library) \u00b7 Role\u2194dimension skipped (dimension not under chosen role)",
        "role_dimension_saved": false,
        "roles_from_db": [],
        "skill_dimension_saved": true,
        "skill_id": 1369,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "Performance and Scalability Tuning",
          "id": 11,
          "rationale": "Techniques for improving throughput, latency, and resource efficiency in backend services. Focuses on profiling, concurrency, load behavior, and bottleneck elimination.",
          "slug": "performance-and-scalability-tuning",
          "source": "db"
        },
        "dimension_id": 11,
        "input_skill": "Distributed Systems",
        "llm_role": null,
        "matched_chosen_role": false,
        "outcome_line": "New skill saved \u00b7 Existing dimension (library) \u00b7 Role\u2194dimension skipped (dimension not under chosen role)",
        "role_dimension_saved": false,
        "roles_from_db": [
          {
            "display_name": "Backend Engineer",
            "id": 1,
            "rationale": null,
            "role_archetype": "A Backend Engineer designs, builds, and maintains the server-side logic and data handling that power applications and services. They focus on implementing reliable business functionality, integrating with other systems, and ensuring the backend is scalable, maintainable, and observable.",
            "slug": "backend-engineer",
            "source": "db"
          }
        ],
        "skill_dimension_saved": true,
        "skill_id": 1369,
        "skill_tag": "in_db",
        "skipped_reason": null
      },
      {
        "chosen_role_id": 3,
        "dimension": {
          "difficulty_hint": "well_known",
          "display_name": "Cloud Platforms",
          "id": 20,
          "rationale": "Proficiency in major cloud service provider platforms and their core services.",
          "slug": "cloud-platforms",
          "source": "db"
        },
        "dimension_id": 20,
        "input_skill": "Distributed Systems",
        "llm_role": null,
        "matched_chosen_role": true,
        "outcome_line": "New skill saved \u00b7 Existing dimension (library) \u00b7 Role\u2194dimension saved",
        "role_dimension_saved": true,
        "roles_from_db": [
          {
            "display_name": "Backend Engineer",
            "id": 1,
            "rationale": null,
            "role_archetype": "A Backend Engineer designs, builds, and maintains the server-side logic and data handling that power applications and services. They focus on implementing reliable business functionality, integrating with other systems, and ensuring the backend is scalable, maintainable, and observable.",
            "slug": "backend-engineer",
            "source": "db"
          },
          {
            "display_name": "Cybersecurity Engineer",
            "id": 5,
            "rationale": null,
            "role_archetype": null,
            "slug": "cybersecurity-engineer",
            "source": "db"
          },
          {
            "display_name": "Data Engineer",
            "id": 2,
            "rationale": null,
            "role_archetype": null,
            "slug": "data-engineer",
            "source": "db"
          },
          {
            "display_name": "DevOps Engineer",
            "id": 10,
            "rationale": null,
            "role_archetype": null,
            "slug": "devops-engineer",
            "source": "db"
          },
          {
            "display_name": "ML Engineer",
            "id": 3,
            "rationale": null,
            "role_archetype": null,
            "slug": "ml-engineer",
            "source": "db"
          }
        ],
        "skill_dimension_saved": true,
        "skill_id": 1369,
        "skill_tag": "in_db",
        "skipped_reason": null
      }
    ],
    "new_skills_created": 9,
    "role_dimension_saved": 0,
    "skill_dimension_saved": 11,
    "skipped": 0
  },
  "planner_output": null,
  "run_id": "2f8955ed-9a4c-4f79-8590-412144b7d135"
}