I’m a Principal Data Scientist at Alior Bank. For 10+ years I’ve built machine learning models in settings where „it works” is a claim you have to prove – to an A/B test, a validator, or an auditor. I started in machine learning, causal inference and experimentation; today I build and evaluate LLM and retrieval systems for Polish-language text.
Selected Projects
Fine-tuning Polish Banking Embedding Models
Alior Bank | Principal Data Scientist | 2026
Fine-tuned a family of Polish text-embedding models for the banking domain – from an edge-deployable encoder to a GPU-class model – using PyTorch, sentence-transformers, full fine-tuning, contrastive learning, and synthetic Q&A generation. On internal and out-of-sample benchmarks strongest models outperforms market SOTA outright, and public retrieval benchmarks confirm no catastrophic forgetting – several general-task metrics actually improved.
GenAI Banking Assistant (PoC)
Alior Bank | Principal Data Scientist | 2026
Designed and built a secure conversational AI for regulated banking on Google Cloud: three cooperating Gemini LLMs (intent, scenario handling, safety) with function calling, maximum-level safety filters, and in-chat PII redaction via Cloud DLP. Delivered a config-extensible, containerized PoC (Artifact Registry, Cloud Run, Streamlit) engineered to production-grade security standards.
Searchless Chess: Evaluation of ViT vs. CNN Architectures
Personal Research Project | 2025 | [GitHub] [Dataset]
Architected and trained a Vision Transformer (ViT) engine achieving ~1960 ELO purely through learned positional intuition, without any search (Minimax/MCTS). Conducted a comparative study between CNN, ResNet, and Transformer architectures, optimizing a massive dataset of Stockfish evaluations.
Document Simplification System
Alior Bank | Principal Data Scientist | 2025
Architected a production-grade simplification pipeline using Python, NLTK, and SOTA LLMs. Achieved a ~300x reduction in processing time and operational costs by automating linguistic analysis and text generation for financial documents.
RAG-based AI Agent
KPMG | Executive Data Scientist | 2023-2024
Engineered RAG agent on Azure, utilizing OpenAI models and Python. Developed a custom document preprocessing and chunking pipeline, optimizing answer accuracy through evaluation-driven iterations in a scalable cloud environment.
Large-Scale NLP Feature Engineering
Alior Bank | Advanced Analytics Expert | 2021-2022
Optimized production ML models by extracting signals from unstructured text data using PySpark and scikit-learn. Orchestrated data pipelines on Hadoop/Hive to process massive datasets, resulting in a measurable Gini coefficient uplift and successful deployment in a distributed production environment.
B2B Recommendation Engine
Alior Bank | Advanced Analytics Expert | 2021
Engineered a recommendation system for business clients by benchmarking interpretable baselines against matrix factorization techniques (NNMF). Successfully validated the system through production A/B testing, achieving a ~160% lift in conversion/engagement compared to the control group.
Enterprise ML Development Framework
Alior Bank | Advanced Analytics Expert | 2020
Architected and led the development of a core Python framework to standardize ML modeling during a bank-wide R-to-Python migration. Built on statsmodels, scikit-learn and modern software engineering best practices, the tool automated manual workflows and became the primary modeling standard for multiple departments.
Customer Propensity-to-Buy Modeling
Alior Bank | Advanced Analytics Expert | 2019
Developed a high-impact propensity model by engineering features from unstructured data at scale using PySpark, Hadoop, and Hive. Implemented a multi-stage feature engineering pipeline and utilized Gradient Boosting to predict customer product interest, delivering insights across multiple banking departments.