Broker CALL-E Agent

Inspiration

Real-estate brokers often spend time calling potential customers just to understand basic requirements such as what they want to buy, where they want it, their property type, budget, and preferred configuration. We wanted to turn that repetitive first conversation into a structured lead that a broker can immediately follow up on.

What we built

Broker CALL-E Agent is an AI-powered real-estate phone agent that calls a customer and collects their property requirements through a focused conversation.

The agent extracts structured information including:

• Intent
• Location
• Property type
• Budget
• Bedrooms
• Timeline
• Additional requirements
• Evidence summary

The result is returned as structured JSON so that the broker does not have to manually read through a conversation to understand the lead.

How it works

The application uses CALL-E for the actual phone conversation and LangGraph to orchestrate the workflow.

The flow is:

1. Validate the customer's phone number.
2. Prepare the call objective.
3. Place the phone call through the CALL-E SDK.
4. Wait for the structured call result.
5. Validate the completed result.
6. Store the extracted requirements locally for broker follow-up.

The backend is exposed through a FastAPI endpoint.

Built with

The project was built using Python, FastAPI, LangGraph, Pydantic, and the official CALL-E Python SDK.

What we learned

We learned how to integrate CALL-E into a real application rather than treating the phone call as an isolated demo. We also focused on reliable structured extraction, call-result validation, unique idempotency keys, call IDs for observability, input validation, and safe handling of credentials and phone numbers.

Challenges

The main challenges were making the phone workflow reliable, handling incomplete conversations without inventing missing information, preventing duplicate calls, and keeping credentials and personal phone numbers out of the public repository.

We also had to handle the difference between a call being initiated and a call actually completing successfully, so the application validates the structured CALL-E result before storing it.

Live verification

The application was tested with a real CALL-E call. The agent successfully collected a customer's requirements including:

• Intent: Buy
• Location: Pune, Hinjewadi
• Property type: Flat
• Configuration: 2 BHK
• Budget: 80 Lakh
• Bedrooms: 2
• Timeline: within 3 months
• additional_requirements: Parking required
• evidence_summary: Customer confirmed looking to buy a 2BHK flat in Hinjewadi within 3 months under 80 lakh budget.

The extracted information was successfully returned and persisted by the application.

Built With

• call-e
• calls
• fastapi
• json
• langgraph
• phone
• pydantic
• python
• voice