About
Hi, I am a ML Engineer with 10 years of experience building production system.
I am mainly interested in designing and developing stable, maintainable systems that keep complexity to a minimum. Focused on thinking hard about SW design and where is the area for biggest added company value.
This description of Impact = Agency + Taste resonates with me, here are some examples from my role at GLAMI:
- [Agency] Bussiness teams were unhappy with non-personalized ranking, however the system was 7 year old thousands of lines of PHP that nobody dared to touch. I suggested and volunteered for a complete rewrite to an AWS backed Python service that would allow faster changes and AB testing loop. I worked on the project solo including entire project management, communication with leadership, other teams etc. and delivered the project on schedule after 3 months.
- The system did not need any further rewrites and my team managed to bring over 10% GMV/s in the following years thanks to the much faster iteration loop and improvements that we came up with.
- [Taste] In the last 5 years my team has run over 30 AB tests with 70%+ of them having positive impact on the main company-wide bussiness metric (GMV/s). Improvements to a recommender system are almost impossible to evaluate offline due to the different behavior per user and feedback loops inside the live system. Therefore knowing which changes are likely to lead to improvement is crucial.
Bio:
Recommendation Team Lead at GLAMI (2020 - now)
- designed and built entire Recommender System from the ground up including data pipelines, models, training, evaluation, deployment, monitoring
- 13 countries, millions of items, millions of users, millions of requests per day, 99th percentile of response time ~100ms
- Python, PostgreSQL, AWS, Kibana, Grafana, FastAPI, Flask, Pandas, NumPy, Scipy.sparse, PyTorch
- first production test after 6 months of joining brought +5% GMV/s
- we keep improving the system, whole AI team grew from only 2 of us to 12 people now and I am leading a 3 person team managing personalization of whole GLAMI + many other projects
- responsible for hiring, team budget, prioritization, cross-team communication, close involvement with applying AI across company and exec team
Quantitative ML Analyst at Qminers (2019-2020)
- Using ML for high-frequency trading at a major Czech Quant firm
University studies (2013-2019):
- Worked part-time at Recombee as ML Engineer & Researcher (2015-2019)
- solo-implemented a production system for Online Optimization of hyper parameters using Gaussian Mixture Modeling and CMAES
- still used in production 10 years later
- implemented and trained novel hybrid Denoising Autoencoders and many other models in TensorFlow (PyTorch was released in 2015 :D )
- solo-implemented a production system for Online Optimization of hyper parameters using Gaussian Mixture Modeling and CMAES
- Founded Let’s Talk ML Prague club in 2015 where students presented hot research papers every week
- I have made 30+ talks explaining new papers
- 2014-2019 was an amazing time with new advances coming out every week and nobody knew which ones are going to last, Adam 2014, GANs 2014, YOLO 2015, ResNet 2015, Distillation 2015, Capsules 2017, Transformers 2017 …
- Bachelor and Master degree from FIT CTU in Prague, both summa cum laude
- Bachelor thesis [EN]: Optimization of Recommender Systems
- Master thesis [EN]: Deep Latent Factor Models for Recommender Systems