Aangan Studio: phone enquiry system

A voice agent (Vaani) answers enquiry calls for Aangan Studio, an interior design studio in Pune. When a call ends, Vaani sends the call to this app, which reads it with OpenAI, applies the qualification rules, saves it in Neon, books/looks up the Cal.com consultation, creates a HubSpot deal for qualified enquiries, emails the designer (Resend) and shows everything on a password-protected dashboard.

Phone calls only. WhatsApp and web forms are out of scope.

How it fits together

• agent/system-prompt.ts — what the voice agent says (includes the one allowed pricing reply). npm run agent:create pushes it to Vaani.
• app/api/vaani-webhook/route.ts — receives the finished call (needs ?token=WEBHOOK_TOKEN).
• lib/extract.ts — OpenAI turns the transcript into plain facts.
• lib/qualification.ts — the rules. Edit rules here; defaults are marked DEFAULT.
• lib/cal.ts, lib/hubspot.ts, lib/handoff.ts — Cal.com, HubSpot, Resend.
• app/page.tsx — the dashboard. It is open to anyone with the link; phone numbers are masked.
• db/schema.sql — the one calls table. npm run setup:db creates it.

Run it yourself

1. npm install
2. Copy .env.example to .env.local and fill in every value. Never commit .env.local.
3. npm run setup:db, then npm run dev.

Tests

• npm run test:rules — rules against 20 scripted calls (no internet).
• npm run test:extraction — the same 20 calls through OpenAI (uses your OpenAI key). The 20 calls are scripted from the project brief, not the original transcripts.

Known gaps

• Vaani does not report time-to-answer or (so far) call cost in the webhook, so the dashboard shows "Not available" for those.
• Vaani's booking tool can report an error even when Cal.com made the booking. The app checks Cal.com after the call.
• Resend without a verified domain only delivers to the account owner's address.