Back to home

ReachScout: evidence-backed professional research

Finds the right recruiter, hiring manager or founder from public sources, shows the evidence behind every data point, and drafts a context-aware introduction. Built to be explicitly not a scraper or spam tool.

ReachScout: evidence-backed professional research

Type

Personal project

Role

Full stack, solo

Duration

Source on GitHub

Tools & Technologies

TypeScriptNext.js 15SupabasePrismaClaude APIVitestPlaywright

Problem

Finding the right person to contact at a company, and having something genuine to say, takes a lot of manual searching. Most tools that automate it scrape private data or guess email addresses and present the guesses as fact.

What I built

A research assistant with a five-stage search pipeline:

  1. Normalize the organisation name
  2. Generate queries, stripping dangerous search operators
  3. Retrieve publicly available results through a swappable search provider
  4. Extract professional profiles, keeping the source for each data point
  5. Rank and deduplicate, merging profiles that describe the same person

Each contact gets a 0 to 100 confidence score with a plain-English explanation of which public source supports which fact. Outreach drafts are generated with Claude, with templates as a fallback when no API key is set.

Honesty built into the data model

  • Emails are always labeled as publicly listed, verified, general contact, or pattern estimate
  • Pattern estimates are hidden by default, never marked verified, and excluded from CSV exports
  • A suppression list records "do not contact" requests with a reason
  • Nothing is sent automatically

Security

  • Every outbound URL is checked against a private-network blocklist before fetching (SSRF protection)
  • Supabase row-level security on all user data, plus ownership checks on every API route
  • Per-user rate limits on searches, profile lookups, verifications and exports
  • No API keys or email bodies in logs

Testing

Unit and integration tests with Vitest and end-to-end tests with Playwright cover the cases that matter most: duplicate profiles merge, official emails are labeled correctly, pattern estimates are never labeled verified, private-network URLs are rejected, and invalid input is refused. A demo mode with fictional contacts on example.com domains runs without any API keys.