A bit more context

I'm an AI research engineer at YoyoAI in Bengaluru, working on speech processing, NLP and generative AI. Most of what I do sits in the gap between a model that performs well on a benchmark and a system that keeps performing once real, messy data is flowing through it at scale.

Day to day that means eight-hour recordings from retail shop floors: finding the customer conversations buried inside them, separating speakers who were never enrolled, and deciding where one interaction ends and the next begins. The work has gone from research prototype to a deployed product running across more than 2,000 stores.

I came to this through a Master's in Computer Science, where I spent most of my time on deep learning and interpretability — how you justify a model's output to someone who has to act on it. That interest stuck. I care about evaluation design, reproducible pipelines, and knowing what a system does when it fails rather than only when it works.

I write up what I learn here, partly to think it through properly and partly because the explanation I wanted usually didn't exist.