What
Broaden accepted inputs beyond books: academic papers, internal wikis/Confluence exports, and video transcripts (YouTube/Udemy/conference talks).
Why
Expands the addressable market well beyond published books, into the much larger space of professional/learning content. Same extraction → generation pipeline, new sources = more use cases with modest effort.
How
- Add input adapters: transcript fetch/parse (VTT/SRT, YouTube transcript API), paper handling (already PDF, but add section/abstract/reference awareness), wiki/Markdown bundles.
- Normalize each into the existing
full_text.txt + metadata.json contract so the generation half is unchanged.
- For transcripts, add timestamp anchors as the "page" equivalent for citations.
Acceptance criteria
- At least one new input type (e.g. video transcript) produces a valid skill.
- Citations use the source-appropriate anchor (timestamp/section).
What
Broaden accepted inputs beyond books: academic papers, internal wikis/Confluence exports, and video transcripts (YouTube/Udemy/conference talks).
Why
Expands the addressable market well beyond published books, into the much larger space of professional/learning content. Same extraction → generation pipeline, new sources = more use cases with modest effort.
How
full_text.txt+metadata.jsoncontract so the generation half is unchanged.Acceptance criteria