Current role · Under NDA
Voice AI agents, built from the ground up.
I work on voice AI agents that hold real, spoken conversations. The work runs from the models the agents are built on to the backend that keeps them responsive under load. Details here are kept general to respect an NDA.
This work is under NDA. Client names, proprietary detail, and product screenshots are left out on purpose. Happy to talk through more in a conversation.
The problem
A voice agent has to listen, think, and reply fast enough that the conversation still feels natural. That means stitching speech recognition, a language model, and speech synthesis into one path without adding the kind of delay a caller would notice.
What I do
I integrate the speech and language models the agents run on, then build product features on top of them. A lot of the day to day is keeping the path from audio in to audio out tight, and making the backend hold up as traffic grows.
The outcome
The agents handle live conversations in production. The work has stayed focused on keeping responses quick enough that the conversation feels natural, and the backend steady as more people use it.
What I owned
- Integrate speech-to-text, language, and text-to-speech models into one real-time pipeline
- Ship new agent capabilities into the product
- Keep the backend fast and reliable as usage scales
Stack