Voice agent platform
The core platform that lets AI agents hold live voice conversations at Nugget by Zomato: audio in, a decision, audio out, inside a strict latency budget.
I built and own the voice agent platform at Nugget by Zomato. It is a distributed, microservices-based system that runs a real-time conversation from the first audio frame to the last word of the reply.
Much of the work sits in the audio path. Voice activity detection, noise cancellation, audio normalisation and turn-boundary detection decide when a person has started and finished speaking. Get that wrong and the agent talks over people or sits in silence.
Around it are the streaming pipelines for live audio ingestion, agent reasoning and response synthesis, and the orchestration layer that handles agent coordination, state, concurrency and failure. The constraint everywhere is the same: stay reliable inside a latency budget, at production volume.
I joined as an SDE working on backend and AI agents in April 2025 and was promoted to SDE-2 in January 2026 for owning this system end to end.
- Role
- Owner, SDE-2
- Since
- April 2025