01Home02Architecture
AI-Powered Voice Agents
Temporarily offline

Velox AI is takinga short break

Velox runs a real-time voice pipeline on AWS EC2. Keeping it live has already burned through the free tier on two AWS accounts — and since Velox isn't generating revenue yet, the live demo is paused while I sort out hosting.

I'm actively working on bringing it back up. In the meantime, here's a recorded demo of Velox in action. 👇

Demo

Trouble loading? Watch the demo on Google Drive

Real-time voice10+ STT / TTS enginesRAG-poweredMultilingual
Go deeper

Curious how it actually works?

After watching the demo, you can understand exactly how real-time voice agents work under the hood — I wrote two deep-dives on building them.

Anatomy of a Real-Time Voice Agent

How a real-time voice agent is structured, end to end.

harshilmistry.com/blogs

Technical Implementation of a Voice Agent

Building the voice pipeline — down to the technical details.

harshilmistry.com/blogs

Powered by a best-in-class stack

Next.js
FastAPI
Deepgram
ElevenLabs
Groq
Cerebras
NVIDIA NIM
MongoDB
Qdrant
Redis
LangChain
Sarvam AI
AssemblyAI
Docker
Next.js
FastAPI
Deepgram
ElevenLabs
Groq
Cerebras
NVIDIA NIM
MongoDB
Qdrant
Redis
LangChain
Sarvam AI
AssemblyAI
Docker

Built end-to-end, solo.

A full real-time voice-agent platform — speech, language models, RAG and telephony, wired together. Peek under the hood while the live site is on its break.

Explore the architecture