F1 Strategy Manager v0.11.0

F1 Strategy Manager v0.11.0

Multi-agent system complete + RAG regulation layer. Seven specialized agents coordinate under a Strategy Orchestrator to produce real-time pit strategy recommendations from live race data.

New: Multi-Agent System (N25–N31)

Sub-agents:

• N25 — Pace Agent: XGBoost lap time predictor with bootstrap confidence intervals → PaceOutput
• N26 — Tire Agent: TCN degradation model with MC Dropout uncertainty → TireOutput
• N27 — Race Situation Agent: LightGBM overtake + safety car probability → RaceSituationOutput
• N28 — Pit Strategy Agent: Pit duration quantile regression + undercut scorer → PitStrategyOutput
• N29 — Radio Agent: Two-stream NLP pipeline (RoBERTa sentiment + SetFit intent + BERT-large NER + RCM parser) → RadioOutput
• N30 — RAG Agent: Qdrant + BGE-M3 regulation retrieval → RegulationContext
• N31 — Strategy Orchestrator: Three-layer architecture — MoE-style routing → Monte Carlo simulation → GPT-4o structured synthesis → StrategyRecommendation

New: RAG System

• scripts/build_rag_index.py — FIA Sporting Regulations PDF → BGE-M3 embeddings → Qdrant (2,279 chunks)
• src/rag/retriever.py — RagRetriever + query_rag_tool importable by N31
• Retrieval scores 0.62–0.76 on demo queries; article citations from chunk metadata

Other changes

• GitHub Actions CI: lint (ruff), typecheck (mypy), test (pytest)
• SRP refactors across all agent notebooks
• LangGraph computation graph visualization cells (N25–N31)
• ROADMAP: v0.9.0 src/ extraction plan, FastMCP architecture, voice model roadmap added