Prospect Finder architecture

Profile building, website reading, verification, scoring, and the hand-off to Contacts.

Prospect Finder turns a plain-language description into verified people in Contacts. Gemini proposes companies and reads their websites; everything that ends up stored is checked in code, so a quote, a name or an email is only kept when it is on a page the site reader fetched.

Runtime flow

A description search has Gemini suggest companies; a company lookup skips that step and resolves the names you entered to websites, then reads people from those sites without scoring them. A search runs in slices of about 50 seconds. The page calls the run endpoint again while the result says it is not done, and progress is saved after every batch of three companies, so closing the tab only pauses the search and Resume continues from the saved state.

From description to verified prospects
Rendering architecture diagram...

Implementation map

Frontend

/prospects wizard, run loop with polling, companies and people views, add-to-contacts modal, and api.prospects methods.

Backend

ProspectsController and ProspectsService, the search runner, profile builder, scoring, SiteReader, verification helpers, and the WebsiteSource candidate source.

Data

ProspectSearch, ProspectCompany and Prospect. Unused rows expire after 90 days; adding a person to Contacts clears the expiry.

Reuses

ContactsService.importContacts for duplicates and directories, WatchService for company watches, and the per-user Gemini key and text model from Settings.

Safeguards

Network

The site reader only fetches public http and https addresses on ports 80 and 443, re-checks every redirect, and resolves addresses itself so private ranges are refused.

Politeness

robots.txt is honoured for the ReachConvertBot agent, a failing robots.txt counts as disallowed, and requests to one host are at least a second apart.

Evidence

A criterion only counts when its quote is found in the fetched text. People must be named on a fetched page, and an email must be published there at the company domain and match the person.

Cost

One search at a time per user, ten searches per 24 hours, and a Gemini call budget of three per company plus five.