A local-first tool that turns RFQs and sales documents into structured, review-ready records — extracting fields, flagging missing information and risks, and tracing every value back to its source sentence.
RFQ and sales documents bury the fields that matter — pricing, volume, program, deadlines — in unstructured text, and manual triage is slow and inconsistent.
A static web app that extracts structured fields deterministically, shows the exact source sentence behind each one, flags risks and missing info, and routes documents through an approve/reject review queue.
Shared ES modules (schema, analyzer, review, store) run identically in the browser and in the Node test suite, so the same extraction logic is covered by 23 unit tests. No server, no API keys, no account.
Choosing deterministic, rule-based extraction over an LLM call so results stay explainable and reproducible with zero API cost — the repo documents how LLM-based extraction could be added later as an opt-in.
How much of "document AI" trust comes from traceability — showing the source sentence — rather than the extraction method itself.
Source repository
Back to the portfolio