02Work
Lumen
WhatsApp AI agent for brokers. Real-estate brokers run their whole workflow on WhatsApp, by text or voice.
2026Consulting · end to endNode.js · LangGraph · Gemini · BullMQ · Postgres
- 19
- skill playbooks
- 130+
- Vitest test files
- 0
- writes without approval
- Text + voice
- on WhatsApp
The problem
Brokers live on WhatsApp but their work is scattered: onboarding, RERA checks, inventory, leads and site visits all happen in different places.
What I built
- 01Onboarding, RERA verification, inventory search, leads and site visits, all inside WhatsApp.
- 02A LangGraph tool-calling loop on Gemini, guided by 19 skill playbooks.
- 03Reminders the agent schedules for itself with BullMQ + Redis.
- 04State kept in Postgres checkpoints, so conversations survive restarts.
How it works
- Agent loop
- Skill playbooks
- Approval gate
- Postgres
- BullMQ
Decisions
Approval before every write
The agent proposes; the broker confirms. Nothing changes in their data without a yes.
Checkpointed state
Postgres checkpoints make every conversation resumable and debuggable.
Next project
GTM Suite
AI go-to-market platform