At Quantum AI Foundation, I’m researching offline multi-agent reinforcement learning for traffic signal control, comparing Decision Transformers with BC/IQL baselines and evaluating transfer from CityFlow to SUMO. I’ve completed the experimental campaign, published a pre-registration, and am preparing the statistics and manuscript for a peer-reviewed ITS venue.
I built the PyTorch offline-RL pipeline and a multi-agent workflow in Claude Code, with specialised agents, hooks, policy-as-code guardrails, and human approval gates. The pipeline includes paired multi-seed evaluation, and its GitHub Actions CI fails when the declared test-skip ceiling is exceeded.
Previously, at ICM, I created an Android workload-assessment tool for air traffic control personnel and developed data pipelines for flight plans and geospatial data. At Dom Maklerski BOŚ S.A., I automated FATCA/CRS reporting, reducing preparation from over a month to on-demand generation; earlier, at Accenture Poland, I supported data products, pipelines, and Power BI dashboards.
