Ingesting 100k papers into an LLM wiki: user seeks editorial policy for AI knowledge synthesis

gintrux · reddit · 2026-09-21

A Reddit user proposes using an "LLM wiki" to distill 100k+ IBS papers into a compressed knowledge base for finding overlooked interventions. Key challenge: ingestion requires constant editorial decisions (create/edit/rename pages, record relations), demanding an explicit editorial policy rather than relying on raw model intelligence. Their pipeline: export PubMed lists, download open-access PDFs, OCR to Markdown, then ingest in batches of 25 — one agent per paper in its own worktree, with a reconciliation agent merging diffs at batch end. They're asking the community for shared editorial policies.

Original post →

More from coding & agent

coding & agent channel →