Fleet Engine is an AI system that finds companies that are likely to need your service, researches exactly what's broken inside each one, identifies a real person to talk to, and writes a first message tailored to their specific situation — not a template, not a blast, one message per company grounded in evidence. It reads public job postings as the signal: a company hiring for RevOps, or bringing in a financial controller, or scaling a growth team is showing you exactly what problem they're paying to solve right now. Fleet Engine reads that signal, enriches it, and turns it into a warm outreach in minutes. This is the engineering record of building it across three versions until the infrastructure stopped failing.
Three agents over n8n, Postgres, and Docker on GCP. 1,500 stress-test cycles across six campaigns. Broke 18 times. Every failure documented forensically before the rebuild.
Claude Code as Admiral. ADK on Agent Engine. GCS file handoff. Three Vertex AI RAG corpora compounding across runs. Clay delivery. Same agents, new substrate.
Reply outcomes flow back into the RAG store. A fourth agent uses that signal to improve outreach before delivery. The pipeline learns what converts.