Ömer Yüksel


My Work

Here you'll find selected work from clients, employment, and competitions, along with a list of clients and testimonials.

Jump to clients and testimonials


Projects

AI Services for European Tenders

I optimized and enhanced Mercell's LLM-based summarization systems that extract key information from large and often redundant tender documents, saving suppliers significant time and improving their decision-making.

My contributions led to a ~30% improvement in extraction coverage, enhanced the quality of summaries, and protected the model from regressions.


Reverse-engineering LLM prompts

A machine learning competition by Kaggle where I ranked 11th out of 2000+ as a solo competitor. The challenge focused on predicting the original prompt used to generate AI-produced text.

Succeeding in the competition required both a fine-tuned model to predict original prompts and an understanding of embedding spaces and adversarial inputs. See also the blog post, Kaggle write-up and the fine-tuned model on Hugging Face.


Trading at the Close

A machine learning competition organized by Optiver, where I ranked 15th out of 4400+ participants as a solo competitor, outperforming teams of quants, data scientists, and Kaggle veterans.

The challenge involved predicting stock market behavior during the auction phase, and I tackled it using a blend of domain-specific feature engineering and online model retraining, allowing my model to adapt to unseen market conditions. See also the write-up and the code.


Vehicle Price Estimation System

I refined and streamlined CarNext's pricing system for determining fair market values of used cars which combines ML and domain knowledge.

My contributions significantly improved the system's reliability and operational efficiency. I developed a monitoring system for real-time tracking of model predictions and feature quality, reduced the inference API recovery time from hours to minutes during service disruptions, and modernized the Python codebase to enhance maintainability.


Modeling the Dynamics of Trading Systems

While at IMC Trading, I focused on high-frequency trading systems, building models to predict the performance impact of system changes. Using Monte Carlo simulations and probabilistic programming, I was able to accurately forecast the effects of system adjustments on success rates, achieving precision within two decimal points.

The models allowed for more efficient decision-making for infrastructure configuration and long-term investment plans. I also presented parts of this work at PyData London conference.


Clients

Companies I've done contract work for:

What Others Say

[...] I must say that it was always very comforting knowing that Ömer can be absolutely reliable to bring all the best qualities to the table. Being engaged in one of the more complex projects within the company, Ömer was always open to discuss architectural decisions and this was one of the qualities that made working with him even more enjoyable. [...]

Tech Lead at Fenergo


[...] Omer is a brilliant human being and a joy to work with. He was very flexible and always open to find a solution even if the problem statement has changed. One of the very best Python developers I have come across, learnt a lot from him. Also his knowledge related to AWS is very impressive. [...]

Project Tech Lead at Schuberg Philis

See my LinkedIn profile for the full versions.