I built a search engine over the full Epstein document dump (1.4M files) as a live demo for LaSearch, a
search engine I've been working on.
It's full-text with semantic understanding and fuzzy matching — no embeddings, no vectors, no ML models.
Everything runs in the browser, results appear as you type with no noticeable lag on a 1.4M document corpus.
The interesting technical constraint was making semantic-like behavior (finding conceptually related results, not
just exact keyword matches) without the overhead of vector search. Happy to go into the approach if there's
interest.
Demo:
https://epstein.lasearch.app LaSearch (the underlying engine, also available as a desktop app for your
own files):
https://lasearch.app