Lobang property matching agent

The challenge

Property agents already exchange buyer requirements and listings in WhatsApp, but manual matching is messy. A useful consumer workflow had to verify agents, filter irrelevant or unsafe messages, score matches, and protect trust before contact details were shared.

What Grail built

Grail built a headless WhatsApp agent that parses listings and requirements, verifies CEA details, checks relevance and safety, scores matches, asks both sides to confirm, and only then reveals contact details.

Stack used

TemporalPostgresWhatsApp Cloud APIEvolution/BaileysFlueSlack admin

Impact

The prototype demonstrates a WhatsApp-native marketplace workflow with guardrails and double opt-in for contact sharing.

How the workflow runs

Deterministic tools run first for CEA, relevance, safety, matching, and confirmation. The LLM writes user-facing copy from sanitized tool facts.

1. An agent sends a listing or buyer requirement in WhatsApp.
2. The system checks identity and parses the property details.
3. Irrelevant or unsafe messages are filtered out.
4. Matches are scored against current requirements.
5. Both sides confirm before contact links are revealed.
6. Admins can inspect the trace.

Human control

The control points were specific to the workflow, so the agent could speed up the work without silently taking over sensitive decisions.

• Double opt-in before contact reveal.
• Relevance and safety filtering before agent replies.
• Admin trace tooling for review.

What shipped

The implementation centered on these shipped pieces:

• WhatsApp intake flow.
• Agent verification.
• Listing and buyer requirement parsing.
• Relevance and safety filtering.
• Match scoring.
• Double opt-in confirmation.
• Admin and trace tooling.