Roxy AI Assistant
Real-time, full-duplex voice assistant with on-device STT/TTS and LLM-powered dynamic tool use.
The problem
Voice assistants lack real-time, full-duplex conversation with contextual reasoning and dynamic tool use.
The result
Built a real-time voice assistant with on-device VAD, STT, and fast TTS — enabling natural, low-latency conversations with LLM-powered tool use.
How it works
Developed a real-time voice assistant enabling full-duplex conversations with on-device VAD, STT, and fast TTS response. Integrated LangChain + LangGraph agent using MCP tools (FastMCP) for contextual reasoning and dynamic tool use. Implemented a Kokoro TTS pipeline with optional echo cancellation (AEC) for natural low-latency speech output. Built a modular system allowing easy swaps of VAD, STT, TTS, or LLM components for flexible configuration.
Tech stack
- Python
- LangChain
- LangGraph
- Groq
- MCP
- Faster-Whisper
- Kokoro TTS
Tags
- real-time
- voice assistant
- agentic ai
- mcp
- low latency