Live demo · engineering document intelligence
Search your documents by meaning, not keywords
A real sentence-embedding model runs in your browser and searches a set of WPS, PQR, NDE reports and code clauses by meaning. Ask a plain-language question — it finds the right clause even when your words don't appear in it.
Loading the semantic model (~23 MB, first time only)… 0%
What this demo is — and isn't
- Real semantic search: a MiniLM sentence-embedding model (all-MiniLM-L6-v2) runs in your browser via WebAssembly, embedding the documents and your question into 384-dim vectors and ranking by cosine similarity — meaning, not keyword match.
- Everything runs client-side. The model (~23 MB) streams from a CDN on first use and is cached; nothing you type is uploaded.
- The corpus is ~36 synthesized WPS / PQR / NDE-report / code-clause / ITP snippets — realistic but invented, no client data.
- In production this is Cyclone VRAG: real project documents embedded into a pgvector store and searched at scale, not one small in-tab index.
Want this over your real WPS, NDE and standards library?
Book a consult