v0.0.2 — Epic Expansion: Full ML Coverage (39 skills)

🚀 Epic expansion: 28 → 39 skills — full machine-learning coverage

Batch D · Math & Architecture Foundations (+5)

• ml-optimization-methods — convex vs non-convex, first/second-order, constrained optimization
• ml-cnn-vision — convolution inductive bias, ResNet evolution logic
• ml-rnn-sequence — gated RNNs, seq2seq+attention, when RNN still fits
• ml-alignment-rlhf — RLHF three stages, reward hacking, DPO alternative
• ml-automl-nas — search space / strategy / performance estimator triad

Batch E · Applied & Interdisciplinary (+6)

• ml-causal-inference — correlation ≠ causation, Pearl ladder, when you need do-calculus
• ml-explainability-xai — LIME/SHAP, intrinsic vs post-hoc, misuse traps
• ml-federated-privacy — FedAvg, threat models, differential privacy budget
• ml-rag-systems — retrieval-augmented generation pipeline decisions
• ml-multimodal — fusion strategies, CLIP-style alignment, modality collapse
• ml-mlops-deployment — training-serving skew, canary/shadow, drift monitoring

Quality gates maintained

• 351/351 blind tests passing (39 × 9)
• Router expanded to 38 routing entries across 10 workflow stages
• GLOSSARY 105 terms; bilingual README; four-layer knowledge sourcing honestly labeled

Full list: see README.md

v0.0.1 — First Public Release

First public release: 28 ML methodology skills (1 router + 27 specialists), 252/252 blind tests passing, bilingual README. See README.md for full details.