Add advanced local RAG notebook with ChromaDB, FastEmbed, and Ollama

[pm32900]

Features:

• Persistent vector store with ChromaDB
• Fast local embeddings with FastEmbed (BAAI/bge-small-en-v1.5)
• Local LLM generation with Ollama (llama3.2:3b)
• Query decomposition into sub-queries
• Reranking with cross-encoder (ms-marco-MiniLM-L-6-v2)
• Retrieval evaluation: Hit-Rate@5 and MRR@5
• Failure mode demonstrations (out-of-domain, hallucination resistance)

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[kacperlukawski]

Hello @pm32900 Thanks for contributing with a cookbook!

I really liked the idea of query decomposition, and reranking, but in general feel the cookbook is too code-heavy to be merged as it is. We typically try to guide readers between the code cells so they can understand what will happen next. Would you be so kind and modify the notebook so it starts with some introduction describing the target system. and then introduce new concepts one by one? Happy to help with that process!

[pm32900]

Hi @kacperlukawski, thanks for the feedback! I've addressed this in the latest commit, the notebook now has detailed narrative markdown cells before each code section, introducing the concept and explaining the why before the how. The intro also includes a full system architecture diagram. Would love your thoughts!