Fariya Raza
System 01

LeadForge AI

Multi-agent lead research and outreach. 6 coordinated agents research, score and write outreach, with a QA agent that rejects and retries before anything ships.

6agents, one shared state
Hover a step for what it does

Text version: Research → Website intel → ICP score → Opportunity → Outreach → QA gate

Problem

Manual lead research took hours per account and produced inconsistent outreach.

What I built

A six-agent LangGraph system with one typed shared state: Researcher, Website Intelligence, ICP Scorer, Opportunity Finder, Outreach Writer and QA Validator, streamed into a Next.js front end.

Key engineering decisions

  • Typed shared state instead of message passing, so every agent reads and writes the same validated object.
  • Retry-with-correction: the QA agent returns structured feedback and the writer regenerates, capped at two attempts.
  • Provider-agnostic LLM layer so models can be swapped per agent without touching orchestration.

Validation and QA

Pydantic-validated outputs at every agent boundary; pytest coverage over the orchestration graph; SSE streaming so reviewers see progress, not a spinner.

Result

Research, scoring and outreach generated end to end with QA before delivery. 6 agents, one state object, retries visible in the UI.