VoiceRAG

A voice assistant you can actually talk to in real time. It listens, looks up answers from your own documents, and talks back — no typing involved.

VoiceRAG chains speech-to-text, an LLM, retrieval over your own docs, and text-to-speech into one real-time loop. It uses retrieval and tool-calling so answers stay grounded in real content instead of being made up. I built the whole thing modular on purpose, so you can swap in different STT, LLM, or TTS models and compare them without rewriting the pipeline. There's a simple web interface for talking to it and testing each piece on its own.