About
One person, both halves of the job
The strategy call and the code that follows it. From the same person.
Background
Sixteen years in product, and the pattern barely changes: the data is already there, and almost no one is using it. That gap is where I started, and it is still the first thing I go looking for.
It pulled me into building. I co-founded Email Machine, an email marketing platform, and took it from an idea to a product that teams paid for. Then a decade at Seznam, the Czech search engine, as senior product manager for its recommender systems. Millions of users, more than 50 million requests a day, and three peer-reviewed papers at RecSys along the way.
The through-line is the one I sell now: most AI work fails on data and architecture, not on the model. Someone has to make the scoping call and then be there when it gets built. I do both.
Independent now, I take on one or two builds at a time: few enough to stay hands-on with every one.
How I work
- Fixed scope, fixed fee: agreed before the work starts.
- Measured. A success metric and a baseline, or it isn't done.
- One accountable person. You work with me, not a rotating bench.
- Honest about data prerequisites, even when that ends the conversation.
- Handover, not lock-in: code, prompts, eval sets and docs are yours.
- Remote by default; hybrid if you're in the EU and it helps.
- EU-based.
Principles
- The simplest thing that works, then grow it in layers.
- Measure, don't guess.
- Most AI failures are data and architecture failures, not model failures.
- No stopgaps built to be thrown away.
- Say the un-fun parts out loud: evals, data gaps, the parts that won't work.
Credentials
Papers
Three at RecSys.
- “Clickbait Mitigation in an Industrial Recommender System” (opens in a new tab) · INRA @ RecSys 2025 (CEUR-WS)
- “Bypassing the Popularity Bias: Repurposing Models for Better Long-Tail Recommendation” (opens in a new tab) · RecSys 2024, Bari
- “Combining Models for Better User Satisfaction in Video Recommendation” (opens in a new tab) · ORSUM @ RecSys 2020 (CEUR-WS Vol-2715)
Talks
- AI Mellontology 2022, Thessaloniki
- Brno Barcamp 2014
In short
- 16 years in product
- Recommender systems at Seznam scale (50M+ requests/day)
- English and Czech
- EU-based
Client references available on request.