How the desk works
DealScanner is one engine that any investor can point at their own thesis. Two operators run on it today — water/wastewater and healthcare — from the same shared database.
1 · The database is the brain
Every listing is scraped once and stored thesis-neutrally — its full text and raw signals. A listing is never stored twice: the URL is a hard unique key, so the duplicate pile-ups that plagued the old version can't happen.
2 · Your settings are the lens
A cheap global filter drops the obvious non-targets — restaurants, salons — before any AI runs. After that, your own thesis (keywords, size band, geography, flags) decides what surfaces. Edit it in Thesis setup and the board re-ranks instantly, for free, because it's just re-reading data that's already there.
3 · AI where it earns its keep
Keyword matching is literal and instant. When wording is fuzzy — a "specialty contractor" that might be a water business — an optional AI re-judge reads the listing and decides, with a reason. Its answers are cached, so you never pay twice for the same call.
4 · Your votes teach it
Every yes / maybe / no is saved with the full context of the deal at that moment. That's the groundwork for an instinct model that will one day score the way you would — not yet built, but collecting from the first vote.