Problem
services/chat-agent/transcript.py's structured extraction is hardcoded to PM-meeting vocabulary: decisions, action_items, risks (see the EXTRACTION_PROMPT and the three SQLite tables it writes to). None of that maps to studying — a lecture or paper doesn't have "action items," it has concepts and definitions.
Proposal
Add a learning-oriented extraction mode using the same two-phase ingest-then-extract pipeline transcript.py already implements:
key_concepts: [{"term": "...", "definition": "..."}]open_questions: things the material raises but doesn't fully explain — good candidates for follow-up study or flashcards- Optionally keep
decisions/action_items/risks for the PM use case behind a mode flag, rather than removing it outright (see issue for the broader PM-vs-study framing question)
Why
Small, contained change to something that already exists and already works — same pipeline, different prompt/schema. High leverage relative to effort.
Problem
services/chat-agent/transcript.py's structured extraction is hardcoded to PM-meeting vocabulary:decisions,action_items,risks(see theEXTRACTION_PROMPTand the three SQLite tables it writes to). None of that maps to studying — a lecture or paper doesn't have "action items," it has concepts and definitions.Proposal
Add a learning-oriented extraction mode using the same two-phase ingest-then-extract pipeline
transcript.pyalready implements:key_concepts: [{"term": "...", "definition": "..."}]open_questions: things the material raises but doesn't fully explain — good candidates for follow-up study or flashcardsdecisions/action_items/risksfor the PM use case behind a mode flag, rather than removing it outright (see issue for the broader PM-vs-study framing question)Why
Small, contained change to something that already exists and already works — same pipeline, different prompt/schema. High leverage relative to effort.