Add enhanced RAG implementation with NFL team's core + performance

[TheTopDeveloper]

…provements

• Enhanced answer.py with GTE-small embeddings + cross-encoder reranking
• Advanced query expansion with domain-specific knowledge
• Dynamic K-value selection based on query complexity
• Context compression for better focus
• Enhanced metadata extraction with 15+ fields
• Smart context formatting with relevance highlighting
• Expected 2-3% additional improvement over NFL team's baseline

[TheTopDeveloper]

Week 6 key takeaways Objective: Fine-tune frontier LLM (GPT-4o) to beat baseline price prediction. Data: Robust pickle loader with fallbacks; validation split honored. Methods: Prompt engineering, OpenAI fine-tune jobs, JSONL prep, eval via MAE/RMSE/MAPE. Results:Baseline: $133.22 → Fine-tuned:$65.66. Learnings: Prompt quality and clean targets matter; small curated datasets can outperform naive baselines; deterministic testing avoids index errors.