feat: upgrade evaluation pipeline with W&B and update docs

[Western-1]

📝 CHANGELOG

All notable changes to the Talk to Your Docs RAG System.

[3.1.0] - 2026-01-01 - W&B Evaluation Pipeline

✨ New Features

• Weights & Biases Integration: Added full support for experiment tracking.
  • New script evaluation/track_experiment.py.
  • Logs Ragas metrics (Faithfulness, Precision, Recall) to W&B cloud.
  • Logs detailed pandas DataFrames with Q&A pairs for analysis.
• Auto-Ingestion (Cold Start): The evaluation pipeline now detects if Qdrant is empty.
  • Automatically generates a synthetic PDF (test_data_autogen.pdf) using ReportLab.
  • Ingests data and cleans up automatically (Zero-setup testing).
• Robust Evaluator Class: Introduced RAGWandbEvaluator class for cleaner, modular evaluation logic.

🔧 Improvements / Performance

• Groq Rate Limit Handling: Implemented a "Monkey Patching" mechanism for ChatGroq.
  • Intercepts invoke calls to enforce a 10s delay between requests.
  • Prevents 429 Too Many Requests errors on Free Tier (8k TPM limit).
  • Uses object.__setattr__ to bypass Pydantic validation on LangChain objects.
• Clean CLI Output: Silenced noisy loggers (httpx, groq, httpcore, qdrant_client) during evaluation.
• Increased Resilience: Updated RunConfig with max_retries=10 and timeout=600s for long-running evaluations.

🐛 Bug Fixes

• Pydantic Validation Bypass: Fixed ValueError: "ChatGroq" object has no field "invoke" by using direct attribute setting.
• LangChain Prompt Handling: Fixed Chain invocation failed error in src/rag.py.
  • Added check: if isinstance(lc_prompt, str) to convert string prompts from Langfuse into ChatPromptTemplate.
• Git Hygiene: Updated .gitignore to strictly exclude wandb/ local directories and artifacts.

📚 Documentation

• README.md:
  • Added Evaluation & Tracking section.
  • Added comparison tables for "Baseline" vs "Tracked" results.
  • Added W&B integration screenshot.
  • Added make track command documentation.
• New Assets: Added docs/rag-eval-metrics-wandb.png.

📂 Files Changed

• Added:
  • evaluation/track_experiment.py
  • images/rag-eval-metrics-wandb.png
• Modified:
  • src/rag.py (Fixed prompt template type error)
  • .gitignore (Added wandb rules)
  • README.md (Added evaluation docs)

---

Upgrade Steps for Evaluation

1. Install new dependencies:

   pip install reportlab wandb

2. Run tracked experiment:

   make track