Reading a paper means opening a PDF, then hunting separately for the code, the citations, whether it replicated, and what genes/drugs it touches. I spent a few months trying to fix that.
Two parts:
-
The map: I embedded 8.5M papers (arXiv, PubMed Central, bioRxiv, medRxiv), ran UMAP to lay them out in 2D, and render them with a WebGL scatterplot. Every dot is a paper — click it for an LLM TLDR, key findings, citations, peer reviews (where they exist), and similar work. Zoom in and the clusters are labeled by topic.
-
Rich paper pages + an MCP server: each paper is rebuilt with linked genes/proteins/diseases/drugs (normalized to real IDs via PubTator, not regex), clinical trials, 3D structures, code, and a citation graph. It's all exposed as an MCP server too, so you can point Claude or an agent at 8.5M papers instead of pasting PDFs.
Stack: Next.js, Postgres (Neon), Cloudflare R2 for content, embeddings + UMAP/HDBSCAN for the map, a WebGL renderer. The hard part was the ingestion pipeline — 45+ heterogeneous sources normalized into one schema — and keeping entity links accurate, since papers use names not accession IDs (so curated link tables + PubTator beat regex every time).
Free, no signup to explore. I'd love feedback on the map UX and whether the TLDRs/entity links hold up for papers you know well.
Known limits: Google indexing is still slow so title-search isn't great yet, some older arXiv figures are missing, and peer-review coverage is partial.
Comments URL: [https://news.ycombinator.com/item?id=48840207](https://news.ycombinator.com/item?id=48840207)
Points: 1