MIKE VIDAL
2025–26·Solo — design, build, ship

Autonomous B2B intent monitoring + outreach — a multi-stage LLM pipeline with human-in-the-loop approval.

homeworksonar
In active build
Year
2025–26
Role
Solo — design, build, ship

Overview

Pings public channels — App Store reviews, Reddit, Shopify Community, job boards — for high-intent signals, classifies leads with a Sonnet-4.6 classifier (tool-use + structured output + prompt caching), finds contacts, drafts personalised outreach, routes each draft through a Telegram human-approval gate, and sends accepted ones via Smartlead with webhook reply tracking. End-to-end pipeline running unattended; humans only see the approval queue.

Highlights

  1. 01Multi-stage pipeline: scrape → classify → email-find → draft → approve → send
  2. 02Single Sonnet-4.6 classifier with tool-use, structured output, and prompt caching — not a multi-agent system
  3. 03Human-in-the-loop approval via Telegram before any email ships
  4. 04Five source scrapers feeding a normalised lead model in SQLite
  5. 05Next.js 16 approval dashboard with Smartlead-webhook reply tracking

Stack

Node.jsNext.js 16Sonnet 4.6 (tool-use, structured output, prompt caching)multi-source ingestionHITL approval (Telegram)SmartleadSQLite
github ↗email ↗·Miami · remote-first