Search 1.4M Epstein files fast – full-text semantic search without AI or vectors

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

How do you represent meaning without embeddings?