[Cookbook idea] RAG over Indian financial regulations (SEBI/RBI) with Claude

[balabommablock-cpu]

Proposed cookbook

A notebook demonstrating how to build a RAG pipeline for Indian financial regulatory documents using Claude.

Why this would be valuable

Indian financial regulation has unique challenges that make it an excellent RAG case study:

1. Cross-referencing circulars - SEBI circulars reference and amend each other (e.g., "as amended by SEBI/HO/IMD/DF2/CIR/P/2021/024")
2. Table-heavy documents - Regulatory thresholds, fee structures, and compliance limits in tables that lose context when chunked naively
3. Amendment chains - Understanding a regulation requires tracing its full amendment history
4. Domain-specific terminology - BFSI jargon that generic embeddings handle poorly

What the notebook would cover

• Parsing SEBI/RBI circular PDFs with structure preservation
• Domain-specific chunking strategies for regulatory text
• Hybrid search (semantic + keyword) for regulatory QA
• Citation tracking back to specific circular numbers
• Evaluation on real regulatory questions

Context

I'm building this as RegAI and would be happy to contribute a cleaned-up notebook version. India's financial services industry is one of the fastest-growing Claude use cases, and a cookbook for this domain would help many developers.

Related

• Existing RAG cookbook covers general use cases well
• This would add a domain-specific, non-English-centric example
• Data sources are all public domain (sebi.gov.in, rbi.org.in)

[darthcoder]

You can only parse something that has some meaning.