Back to Projects

Speech-to-Speech LLM Agent

Independently built a CLINKS Corporation proof of concept for speech-in and speech-out LLM interaction.

Overview

At CLINKS Corporation, I independently scoped and built a proof of concept for a voice agent. Speech Recognition, an LLM, and speech synthesis make up the conversation loop.

I used FastAPI and WebSocket in the prototype. I prepared it for internal evaluation and stakeholder demos.

What I Built

  • Speech-to-Speech

    I connected Speech Recognition, an LLM, and speech synthesis in one conversational loop.

  • Independent Build

    I scoped the proof of concept and implemented it independently.

  • Working Prototype

    I built a prototype to test the core voice interaction loop.

  • Demo Preparation

    I prepared the prototype for internal evaluation and demos.

Problems

  • Keeping latency low in a real-time voice loop.
  • Handling recognition errors without losing dialogue context.
  • Balancing robustness with the pace of a proof of concept.

Results

  • Built a functional speech-to-speech agent prototype.
  • Tested the architecture for conversational voice interaction.
  • Prepared the prototype for stakeholder demos.