01 / 05 Case study · FourDevs.ai
Sandstorms AI
A voice-agent SaaS for inbound and outbound calls. Businesses put a human-sounding AI agent on their phone line to answer questions, run outbound campaigns and book appointments. I built it alone, end to end.
- Role
- Solo developer, end to end
- When
- Feb – May 2026
- Built at
- FourDevs.ai
- Status
- sandstorms.ai
01 — Problem
What needed solving
Most voice bots feel slow because they work in turns: wait for the caller to finish, write the whole answer, then turn it into speech. On a phone call that pause is long enough for people to talk over the bot or hang up. A business also needs more than a demo: phone numbers, a knowledge base, booking, billing and somewhere to manage it all.
02 — Build
What I built
- A fully streaming speech-to-text → LLM → text-to-speech pipeline, so the agent starts speaking before the full answer exists.
- Barge-in, so callers can interrupt mid-sentence and the agent stops to listen.
- Phone integration over Twilio and SIP, for inbound calls and outbound campaigns.
- Voice cloning, a knowledge base the agent answers from, and calendar booking.
- The SaaS around it: subscription billing, multi-tenant roles and an admin console.
03 — On a call
What callers hear
- Caller
Hi, can I move my appointment to Friday?
- Agent
Friday at 11 is free. Shall I move it there?
- Caller
Actually, make it two o’clock.
Barge-in · agent stops - Agent
Done: Friday at 2 pm. It’s on the calendar.
04 — Architecture
How it works
05 — Outcome
Results
- Handles live calls in production.
- Sub-second time to first audio.
- Callers can cut in naturally instead of waiting for the agent to finish.
06 — Stack
Built with
- React
- Node.js / Fastify
- PostgreSQL
- Deepgram
- Groq
- ElevenLabs
- Twilio