Velox AI.
A B2B SaaS platform for building and deploying real-time voice agents through a no-code dashboard — sub-700ms latency, with first-class support for ten Indian languages.
01Overview
I built Velox AI for the people who own the outcome of phone interactions but don't write code: founders, ops leads and customer-experience managers. A seven-step Agent Builder turns a handful of form fields into a working, deployable voice agent in under five minutes.
Under the hood it's a streaming-first system. Audio runs through a four-task asyncio graph — STT loop, LLM stream, TTS worker, audio sender — coordinating partial transcripts, sentence-buffered speech and mid-sentence interruption. The defining choice is a provider-agnostic adapter layer: STT, LLM and TTS vendors are swappable per agent from the dashboard.
02Tech stack
Frontend
Next.js 15 · React 19 · TypeScript · Tailwind CSS
Backend
Python 3 · FastAPI · WebSockets
Data
MongoDB · Redis · Qdrant
AI & ML
Deepgram · Sarvam · ElevenLabs · Groq · Cerebras · NVIDIA NIM
Infra
Docker · AWS EC2 · Nginx
03Highlights
- 01
Sub-700ms round-trip from user silence to agent reply
- 02
Full-duplex WebSocket streaming with semantic barge-in
- 03
Swap STT, LLM and TTS vendors per agent
- 04
Native support for ten Indian languages
- 05
Per-agent RAG knowledge base on multi-tenant Qdrant
( Keep reading )
