Work

Much of my work has been on what sits behind an AI product: how data reaches a model, how computation scales, and what happens when a system fails.

Building a platform around the models

At Kumo.AI, I helped build the data and ML infrastructure from scratch. We grew from a single node to a distributed platform training hundreds of models a day.

I worked on the path from customer data to model execution: native Snowflake and Databricks integrations, distributed compute, and a feature store built on RocksDB to support graph neural networks.

Working on systems at scale

On the Google Ads databases team, I worked on distributed storage and led a migration from in-memory lookups to centralized storage with RDMA caching. I also worked on data processing that could tolerate failures across a data center.

That systems background sits alongside my more recent work on Codex at OpenAI and my current role as CTO at LevelUp Labs.

Writing about production AI

I write about building enterprise AI systems with Aishwarya Naresh Reganti. Our articles cover the AI product development lifecycle and the role of evaluation in improving AI products.

We also joined Lenny’s Podcast to discuss reliability, customer trust, and what teams learn after deploying AI products.