What I do
I'm a mechanical engineering PhD turned machine learning leader, with hands-on experience spanning a wide range of domains including real-time bidding, recommendation systems, fraud detection, NLP, agentic systems and many more. Over the years I've also grown into leadership, guiding ML teams, setting technical direction and owning processes from inception to scale.
What I care about most is the part that often gets overlooked: getting models out of the notebook and into production, and making sure they deliver results that actually move business metrics. A clever model that never ships, or that ships but doesn't change the numbers that matter, isn't finished work. That's why I love to break problems down to their core and build solutions up from first principles.
This site is where I share my thinking on ML, engineering, leadership, hype, and the messy space in between.
I'm always happy to talk shop — especially when the problem is hard and the answer isn't obvious. If you've got one, or just want to compare notes, get in touch.
Selected work
- 2025Improving User Targetting
In this task I went back to basics, analyzed our RTB data and identified segments in which our algorithms were underperforming. I then introduced modifications that lead to a 50% increase in ROAS and ARPU across the network
- 2024Metrics Development
This one seems to follow me around, but as we all know - if you don't know what to measure, you can't improve. Here in one case I've gradually changed CEO-made metrics, to industry standard, leading to an increase in rate of successful POCs. In another I've developed metrics from first principles, allowing us to save significant time and money on failing A/B tests that were saved.
Recent writing
August 9, 2026
When the Perfect Future Becomes Your Biggest Competitor
What Amazon Go, Supersmart, and years of building production ML taught me about reliability, technological hype, and the responsibility to communicate what we build.
July 14, 2026
evaluation in recommender systems - part A
Some algorithms are harder to evaluate then others.
June 12, 2026
how-to-store-secrets-in-an-agentic-world
We need to talk about a better secret storage pattern.
Talk to my agent
An AI assistant familiar with my work. Ask it anything you'd ask me.
Ask anything. The agent will stream its reply.