Building the map / 0:00

Real neuroscience papers from OpenAlex, placed by meaning.

1. Countsize of the OpenAlex pool
2. Fetcha seeded random sample, 100 works per request
3. Embedtitle and abstract become 384 numbers, computed on this machine
4. Relatenearest papers by cosine similarity
5. PlaceUMAP, then fitted into the silhouette
6. Assemblewrite the atlas

The first build takes a few minutes: about 8,000 works are sampled, their abstracts embedded here and the layout computed. The result is cached, so later loads are immediate.