Projects / AI data systems
AI data systems
Pipelines that find, verify and extract business data with AI, from a lead engine to a clinic directory and an acquisition sourcing system.
- Role
- Builder
- When
- January 2026 to now
- Status
- OT Directory in testing
- Stack
- Python, SQLite, Supabase, Claude, Apify, NeverBounce, Google Places, NPI registry, Linux
Lead Forge AI (January to February 2026): business discovery through Google Places and Apify, GPT-4o-mini scoring, NeverBounce verification and Supabase functions. Tuned scraper concurrency, added cost tracking and fixed the scoring rubric. Finished.
OT Directory (summer 2026 to now): a joint venture building a public directory of occupational therapy clinics. The pipeline reuses Lead Forge: NPI registry discovery, Census geocoding, Google Places lookup, site crawl, an ownership check, Claude extraction, duplicate merge and map-pin correction. A first extraction model was replaced after measuring 35% field errors. In testing before launch.
Acquisition sourcing (summer 2026): an AI-assisted pipeline for finding small businesses to buy, built with my dad. 29,245 Google Maps queries, a buy box built on SBA 7(a) default data across 27 industry codes and 8 states, and 590 verified owner contacts. Built and paused before any outreach; no emails were sent.
Hermes: an always-on Ubuntu mini PC with an encrypted disk, Tailscale and systemd, running a local 20B model, Whisper lecture transcription and email triage.