Skip to content

Changelog

All notable changes to this repository will be documented in this file.

[Unreleased]

Added (April 2026 Remediation — Audit Hardening)

  • 3 new frontier notes (Phase 3):
  • applications/ai-coding-agents.md — Cursor/Devin/Windsurf/Codex comparison, minimal agent implementation, production failure modes
  • techniques/structured-outputs.md — Constrained decoding, JSON Schema enforcement, provider comparison (OpenAI/Anthropic/Google), Outlines/BAML
  • ethics-and-safety/mcp-security.md — Tool Poisoning, Rug Pull attacks, OWASP mapping, vulnerability scanner, hardened MCP server
  • All 3 new notes registered in mkdocs.yml, genai.md scope map, and LEARNING_PATH.md
  • CI/CD hardening (Phase 2):
  • scripts/check_depth.ps1 — New quality gate enforcing minimum substantive code depth per difficulty level
  • scripts/verify_repo.ps1 — Added LEARNING_PATH sequential numbering validation
  • lint.yml — Integrated check_links.ps1 and check_depth.ps1
  • deploy.yml — Added pip caching for faster builds

Fixed (April 2026 Remediation — Audit Hardening)

  • LEARNING_PATH.md Track B duplicate numbering bug (B6/B12 appeared twice, B5 skipped)
  • Code depth for 4 flagged notes — all 75 intermediate+ notes now pass depth check (0 thin, 0 missing blocks)
  • inference/distributed-inference-and-serving-architecture.md — Added GPU health monitor, P/D disaggregation router (2→3 blocks)
  • inference/inference-optimization.md — Expanded quantization code with generation test
  • production/model-serving.md — Added health check and async load test endpoint
  • tools-and-infra/ml-experiment-and-data-management.md — Added run comparison and model promotion gate
  • vLLM version references standardized to >=2.0 across all notes
  • Historical inject scripts marked with DO NOT RE-RUN warning headers

Added (April 2026 Remediation - Phase 6)

  • Phase 6: CI hardening - Code & Implementation now a blocking gate
  • scripts/check_content_compliance.ps1: promoted ## ★ Code & Implementation check from informational warning to hard blocking CI gate
  • All 4 compliance dimensions now enforced (exit 1 on any failure): Production Failure Modes, Hands-On Exercises, Code & Implementation, OWASP stale terms
  • CI contract: any new intermediate+ note without a Code section blocks merge
  • Label updated from [non-blocking until Phase 6] to [ENFORCED - Phase 6 gate]

Added (April 2026 Remediation — Phase 4 and 5)

  • Phase 4: Frontier content injection
  • techniques/advanced-fine-tuning.md: RLVR and ORPO sections with 2026-04 ground truth
  • ethics-and-safety/ai-regulation.md: Full overwrite — EU AI Act Aug 2, 2026 deadline, Digital Omnibus, NIST AI RMF, programmatic compliance checker
  • llms/llm-landscape.md: Gemma 4 family (E2B, E4B, 26B MoE, 31B Dense), multi-provider API comparison
  • foundations/transformers.md: Code and Implementation section with HuggingFace inference and PyTorch Transformer block
  • Phase 5: Code section injection (all 72 intermediate+ notes)
  • All notes now have ## ★ Code & Implementation with versioned, working Python snippets
  • 31 notes injected via automated bulk injection scripts (reusable for ongoing maintenance)
  • CI quality gate fixes
  • OWASP stale-term check: fixed false positives on historical comparison tables
  • All 4 compliance dimensions at zero: Production Failure Modes, Hands-On Exercises, Code sections, OWASP stale terms

Added

  • README.md as the public landing page for the repository
  • LEARNING_PATH.md with the Universal Foundation and 5 role-cluster tracks
  • CONTRIBUTING.md with contribution guidelines and note standards
  • CHANGELOG.md for visible version tracking
  • LICENSE using CC-BY-4.0
  • _templates/career_role_template.md for future per-role guides
  • Phase 1 critical notes: production/ai-system-design.md, llms/hallucination-detection.md, agents/multi-agent-architectures.md, agents/agent-evaluation.md, techniques/advanced-fine-tuning.md, production/docker-and-kubernetes.md, production/model-serving.md, production/monitoring-observability.md, production/cicd-for-ml.md, evaluation/llm-evaluation-deep-dive.md, research-frontiers/distributed-training.md, inference/gpu-cuda-programming.md, ethics-and-safety/ai-regulation.md, applications/conversational-ai.md, production/cost-optimization.md
  • Phase 2 career guides: career/roles/ai-engineer.md, career/roles/genai-engineer.md, career/roles/llm-engineer.md, career/roles/rag-engineer.md, career/roles/agentic-ai-engineer.md, career/roles/ml-engineer.md, career/roles/mlops-engineer.md
  • Grouped career guides: career/application-and-strategy-roles.md, career/applied-ml-and-domain-roles.md, career/research-and-infrastructure-roles.md, career/safety-and-governance-roles.md assets/data/topic-role-matrix.json, downloads/anki/README.md, downloads/progress/README.md, generated/topic-role-relevance-matrix.md
  • GitHub Actions workflows: .github/workflows/lint.yml, .github/workflows/links.yml, .github/workflows/deploy.yml
  • scripts/generate_learning_assets.ps1, scripts/check_links.ps1, and scripts/markdown_quality_check.ps1

  • requirements-docs.txt for reproducible docs-site builds

  • repo-readme.md as the MkDocs-site-friendly repository overview page

  • scripts/check_content_compliance.ps1 rebuilt: removed duplicate script body, added Code section reporting and OWASP 2023 stale-term detection
  • scripts/check_freshness.ps1 upgraded to CI-blocking (exit 1 when stale notes exceed threshold)
  • Security note ethics-and-safety/owasp-llm-top-10.md overhauled to OWASP LLM Top 10 2025 — added LLM07 (System Prompt Leakage) and LLM08 (Vector and Embedding Weaknesses)
  • Frontier content injected: Late Chunking + Contextual Retrieval + CRAG in techniques/rag.md; SGLang + P/D Disaggregation in inference/inference-optimization.md; GRPO/ORPO/RLVR in techniques/advanced-fine-tuning.md; Gemma 4 in llms/llm-landscape.md; EU AI Act Digital Omnibus in ethics-and-safety/ai-regulation.md
  • Code & Implementation sections added to 45 notes across all 13 domains
  • mkdocs.yml nav: removed duplicate LEARNING_PATH.md entry, fixed broken repo-readme.md reference
  • CHANGELOG.md: corrected 3 broken file references (training-infrastructure → distributed-training, ml-experiment-tracking + data-versioning-for-ml → ml-experiment-and-data-management)
  • CONTRIBUTING.md: replaced line-count minimums with quality-signal minimums; make is now the primary command
  • index.md: corrected note count from 74 → 78 throughout
  • .github/workflows/lint.yml: added freshness check and link check steps
  • _templates/MAINTENANCE_PROMPTS.md added: session-agnostic maintenance prompt library

Changed

  • genai.md now uses valid related links and the correct career reference link
  • Published note bodies now use standard markdown links instead of GitHub-unfriendly wiki-links
  • Frontmatter parent and related links were normalized across published notes
  • _templates/notes_template.md now reflects the hybrid link policy in its body examples
  • README.md, LEARNING_PATH.md, and genai.md now reflect the expanded Phase 1-4 content footprint
  • LEARNING_PATH.md now explicitly covers specialized electives and adjacent topic paths that sit outside the five main tracks
  • README.md, LEARNING_PATH.md, and CONTRIBUTING.md now surface the docs site, learner tools, and generation workflow
  • Existing overview notes were cross-linked to new production, evaluation, safety, infrastructure, and application notes
  • career/genai-career-roles-universal.md now uses the standardized public repo frontmatter and links to both dedicated and grouped career guides
  • scripts/verify_repo.ps1 now enforces learning-path coverage, genai scope-map coverage, and full code/interview coverage for published topic notes

Notes

  • GitHub repository metadata in the web UI is still a manual follow-up step