Search (semantic)

The Search box (in the top bar, or the dedicated Search page) is semantic, not keyword. It finds estimates by meaning — the content of auditor notes, line items, follow-ups, and the estimate header — not just an exact string match.

How it works

1. Your query is embedded with Workers AI (bge-base-en-v1.5).
2. The vector is matched against the Vectorize index (estimateiq-notes), which holds the full content of every estimate.
3. Results are cached in KV for a few minutes, so repeated queries return instantly (the results show a cached tag when served from cache).

GET /api/search?q=... returns the top matches — estimate, insured, carrier, status, and a match score — each linking to the estimate’s detail page.

Example queries

• water mitigation moisture readings
• O&P dispute
• roof code upgrade
• depreciation holdback

Keeping the index fresh

The index is (re)built by POST /api/admin/reindex, which embeds every estimate’s full content into Vectorize. Run it after a bulk data load or a large batch of updates. See Data loading.