The problem
What was getting in the way
Businesses lose leads and customers because nobody has time to call every lead, follow up on time, or answer every incoming call.
What I built
What it does
- Import leads from CSV/Excel or a single Google Maps link; AI organizes messy files automatically
- Researches each business on the web before calling and drafts a call plan for that lead
- Real phone calls with a real-time two-way AI voice that follows a persona, script and campaign goal, in multiple languages
- Calls through Twilio or through the business's own SIM card, using an Android gateway app I built
- AI receptionist answers incoming calls, takes messages and emails a summary
- Every call recorded, transcribed and summarized, with the outcome extracted (interested, callback, meeting, not interested)
- Listen live, guide the AI mid-call, or take over the call and hand it back
- Automatic retries for failed calls, with a plain-English reason for each failure
- "Call this person, ask this, tell me what happened" tasks with results by email or WhatsApp
- A voice and chat assistant that runs the app on command
- Incoming calls ring the team in the browser first, then the AI answers if nobody picks up
- Installable app with push notifications, and a usage-and-cost dashboard for every AI feature
How it's built
Under the hood
A Node.js app with a React front end. Phone audio streams in real time between the phone network and Google's Gemini Live voice model. A job queue runs research, calling and follow-ups unattended.
React 19
TanStack Start
Node.js
MySQL/MariaDB
Twilio
Google Gemini Live
Kotlin (Android)
Proof
Results so far
- Live at growthintellab.com
- First real campaign: 697 leads imported from one file; about 75 real calls placed in a single afternoon, unattended
- The AI replies about 2 seconds after the person stops speaking (median, measured on real calls)
- The AI receptionist has answered real incoming calls end to end, with transcript and summary saved
Proves these services
Screenshots