AI front office for a real-estate brokerage

The problem

A brokerage lives and dies by inbound calls — but agents are in showings all day, and every missed or fumbled call is a buyer who dials the next office on the list. Reception can answer, but can't qualify a buyer, calm a frustrated seller and book a viewing at 9 PM on a Sunday.

What we built

• A multi-agent squad with warm handoffs. Ava answers every call, figures out what the caller needs, and passes them to the right specialist — Mike for buying and renting, Sarah for selling and landlords, Tom when someone insists on a human. Transfers are announced, and nobody ever asks you to repeat yourself.
• A live CRM board you can watch. Every call ends as a card on a Notion pipeline board — name, intent, budget, area, booked viewing or waitlist spot, recording and summary. Make a test call and watch your own card appear.
• AI post-call analysis and follow-up. After each call, GPT-4.1 reviews the transcript — what went well, what to fix, how warm the lead is — and sends the caller a personalised follow-up email.

Try it

Under the hood

Caller ── browser web-call ──▶ Vapi squad (4 assistants, warm handoffs)
                                  │ STT Deepgram Nova 3 · LLM GPT-4.1-mini · TTS Deepgram Aura
                function tools ───┤
                                  ▼
                     Cloudflare Worker ──▶ D1 (leads · calls · inquiries)
                       │ end-of-call report
                       ├─▶ GPT-4.1 call analysis ──▶ personalised follow-up email
                       ├─▶ Notion CRM board (live card per call)
                       └─▶ Telegram alert to the team
      

The backend is a single Cloudflare Worker: it answers the agents' tool calls in milliseconds during the live conversation, then does the slow work — analysis, CRM card, email, alerts — after the call ends. Watch the CRM board — your test call will appear there.

Want this for your business?