v2.2

New Features:

1. RAGAS Evaluation
   • Added dedicated evaluation.ipynb notebook to evaluate retrieval and answer quality with RAGAS metrics.
   • Evaluation outputs can be saved and reused without rerunning the RAG pipeline.
   • Added support for a separate local judge model through Ollama's OpenAI-compatible API.

2. Updated Retrieval Defaults
   • Switched the default embedding model to Qwen/Qwen3-Embedding-0.6B.
   • Updated retrieval defaults to threshold 0.4 and k=7.

Bug Fixes:

1. Project Cleanup
   • Cleaned dependencies, generated files, and notebook outputs.
   • Updated documentation to reflect the current evaluation, retrieval, and installation setup.