AI researcher · builder · fourth-year student at AGH
I study how intelligent systems make decisions - and build the ones that remove real work.
I turn real problems into models and systems that make people’s work simpler. I combine research with implementation - and test whether the result genuinely helps.
- Data science
- Medtech
- Fintech
- Automation
Selected work
Research, products, and the evidence behind them.
I’m still early in my career, so I would rather show the work than inflate the title. These projects span healthcare, retail operations, and causal decision analysis.
01Medical NLP · Research2026
Best Paper Award · PP-RAI 2026
BloodAI
Multi-label triage for hematological comorbidities using BERT.
Together with Jan Banasik, I developed a clinical routing pipeline that combines basic laboratory parameters with a short adaptive interview. A medical-domain BERT model outputs independent probabilities for eight care routes instead of forcing one label on patients with comorbidities.
Reported held-out validation
ROC AUC / 8 routes
Primary care0.98
Gastro0.92
Haematology0.94
Nephrology0.95
Emergency1.00
Cardiology0.89
Pulmonology0.75
Hepatology0.91
325,829train + validation encounters
8independent care routes
2MIMIC + Synthea data sources
Research prototype for clinician decision support - not a diagnostic system.
02Retail optimisation · ProductLive
Built for a real purchasing workflow
Procero
One place for purchasing, supplier comparison, and inventory decisions.
I built Procero for small and medium grocery chains. It brings sales, inventory, and wholesaler price lists together to recommend what to order, when, and from whom - without managers manually comparing dozens of files.
PROCERO / CASE STUDY9-store grocery network
Time to prepare an order
2 h→8 min
Annual purchasing≈ PLN 50M
Reported price savings3-4%
Wholesalers compared18
Estimated annual value≈ PLN 1.5M
Figures reported in Procero’s published customer case study.
03Causal inference · ResearchIn review
Details limited during double-blind review
Causal decision analysis
Testing when observational action valuations are too fragile to trust.
I’m studying how conclusions change across defensible choices of estimand, nuisance model, trimming, covariates, and feature definitions. The work connects causal inference with real decision logs in sport and other high-context domains.
Research scope
Double-blind review in progress
Conclusion sensitivity ↑
Analyst-choice variation →
01known-truth benchmark validation
02multi-cohort football evaluation
03cross-domain stress testing
Full title, contribution, and results will be added after the review process.
In progress
What I’m working on now.
Agents / RAGKnowledge and workflow agents for the window industry
A current industry project connecting domain knowledge with agent workflows. Details and evaluation will be published when the system is ready to show.
Quant / Causal modellingInformation Diffusion Dynamics Network
A research concept asking whether the path and topology through which information reaches market participants causally affects price dynamics - independently of content.
01Agent workflows and RAG
I connect language models to company knowledge, tools, and existing processes - then evaluate whether the workflow is actually reliable enough to use.
Workflow map · working prototype · evaluation02Process automation
I look for repetitive decisions and manual hand-offs that can be removed with software, data, or an agent - not AI for its own sake.
Process review · automation · integration03Applied AI research
I turn a technical question into experiments, compare credible methods, and document what works, what fails, and what the evidence supports.
Research review · experiments · recommendation About
I’m Antek. I like learning hard things and making other people’s work easier.
I’m a fourth-year Computer Science and Intelligent Systems student at AGH University of Krakow. I want to grow into an AI researcher who can move comfortably between mathematics, code, and the reality of a business process.
I’m naturally interdisciplinary. Medical NLP taught me to think about safety and validation. Causal inference made me more careful about conclusions drawn from data. Procero taught me that a technically interesting system still has to fit the way people actually work.
Outside the projects themselves, I enjoy hackathons, networking, and connecting people who should know each other. I volunteer in educational initiatives and I’m currently building a Quant Finance Club at AGH. I try to work hard, think clearly, and leave a process better than I found it.
Domain Group Lead
Athletic Intelligence
Leading algorithm work for football clubs.Data Science & AI
E-commerce
Working on applied data and automation problems.Former AI Researcher
AI Lab, AGH
Student research in artificial intelligence.Founder
Quant Finance Club, AGH
Building a new community around quantitative research and finance.Recognition
Selected, not exhaustive.