ArxivRagProposal
Jeff Smith edited this page 2026-04-08 17:16:51 -06:00

Page revisions

2 Commits

Author SHA1 Message Date
Jeff Smith
9dcab7627f Roadmap M5.1 → arxiv-rag; resolve proposal open questions
Per session decision 2026-04-08:
- M5.1 was the placeholder file/document researcher; replaced with
  the arxiv-rag researcher (Issue #37). The grep-based file
  researcher is demoted to 'future ideas' since no concrete user
  corpus drove its design.
- All five open questions in the proposal resolved in favor of the
  v1-simplest path: local embeddings (nomic), arxiv IDs only,
  pin paper versions, embed-model in chunk hash, embedding_calls
  ledger field.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-08 17:16:51 -06:00
Jeff Smith
7e5bed6033 Add arxiv-rag implementation proposal
Detailed design proposal for the second researcher: a RAG over a
user-curated arXiv reading list. Locks in chunking, vector store,
embeddings, PDF extraction, and storage layout decisions; lists
open questions and success criteria.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-08 17:08:25 -06:00