User Story

As Maya, in order to get extractions that reference relevant regulations and form instructions, I want an extraction variant that retrieves policy context via RAG before extracting fields.

Context

The homework repo has a working ChromaDB + sentence-transformers RAG implementation. This story brings retrieval-augmented generation into the extraction pipeline: embed form instructions/CFR sections, retrieve relevant context for the uploaded PDF, and include it in the extraction prompt.

Acceptance Criteria

• RAG retrieval primitive (embeddings + vector store) integrated
• Policy corpus seeded (form instructions for the 3 evaluation fixtures)
• New variant extraction/sonnet-with-rag registered
• Evaluation run comparing RAG variant against baseline Sonnet
• Catalog page with findings and course connection
• Roadmap updated

Definition of Done

• Tests pass (bun run check)
• Evaluation complete with LLM judge
• Catalog page documents approach, metrics, and course connection