Lead OPS
This project is a production agentic automation system that handles the full lead intake pipeline from inbound form, email, or Slack message through extraction, classification, CRM update, and team notification. The core design treats confidence scoring and human-in-the-loop review as first-class concerns, not afterthoughts, so every automated action carries an auditable decision behind it. Agents are built on LangGraph or LangChain, connected to external tools via n8n, and grounded with RAG where reliable context is required.
Built end to end from the raw brief, exactly as it arrived. No spec sprint, no Forge blueprint: the floor, not the ceiling.
Lead intake & classification
Proves inbound leads are auto-extracted and classified with confidence scores.
- 1Log in as an ops user
- 2Submit a seeded inbound form/email lead
- 3Watch the agent extract fields and assign a classification with confidence
CRM update & Slack summary
Shows the agent acting on classified leads by updating CRM records and posting a summary.
- 1Open a classified lead
- 2Trigger the agent action to update the CRM record
- 3View the generated summary notification in-app
Low-confidence review queue
Demonstrates human-in-the-loop review for uncertain cases before action.
- 1Open the review queue
- 2Select a flagged low-confidence lead
- 3Approve or correct the classification
- 4Confirm the record updates and clears from the queue