Voice & Memory.
A realtime voice assistant with persistent memory and tools for calendar, CRM, and search.
Python · Whisper · GPT-4o · ElevenLabs · LangChain · Google Cloud Run
The problem
A useful voice agent needs to respond quickly, retain context, and take action through external tools within the same interaction.
The engineering
Connected Whisper speech-to-text, GPT-4o, and ElevenLabs text-to-speech in a realtime voice agent.
Added persistent vector-store memory and multi-tool orchestration across calendar, CRM, and search APIs.
Used asynchronous streaming to reduce latency, then deployed on Google Cloud Run with CI/CD and automated quality regression tests.
The outcome
Achieved sub-1.5-second latency through asynchronous streaming, with persistent memory and tool orchestration in the deployed agent.