About

I am a senior software engineer working on distributed systems and machine learning infrastructure.

I began writing software professionally in 2016. Over the following decade, across e-commerce platforms, marketplaces, and data pipelines, my work concentrated on the problems that appear when one machine is no longer sufficient: event streams that must preserve ordering, caches that must remain consistent, search clusters that must stay available under sustained ingestion, and the services that depend on all of them.

Machine learning introduced the same class of problems in a different form. Training a model is a well-understood process. Serving it to a large number of users, keeping it fast under variable load, and releasing new versions without disruption is where most of the engineering effort lies. Over the last several years I have built real-time serving paths, pipelines that process millions of records a day, and the release tooling that allows a team to ship models on a weekly cadence.

My current focus is inference for models: serving them at scale without idle accelerator capacity, sharing hardware between production and research workloads, and deploying new models with appropriate controls.

This site holds notes on those subjects. The earlier entries were migrated from a WordPress blog I maintained from 2016.