Neuroslop is an independent two-person research shop. We spend most of our time on retrieval quality, evaluation harnesses, and the unglamorous data plumbing that sits underneath both.
Everything we publish is open source. We do not sell a product, run a hosted service, or collect user data.
Chunking strategies and reranking experiments on long, messy technical corpora.
Reproducible harnesses, variance reporting, and honest error bars on small benchmarks.
Dataset linters, trace viewers, and other utilities we needed and could not find.
Notes and code land irregularly — usually when an experiment finishes rather than on a schedule. If something here is useful to you, take it.