MSc Engineering Physics, University of Coimbra. Completed July 2026, final grade 16 out of 20 (Very Good), dissertation graded 18 out of 20.

I work where physical systems meet data: control, simulation and applied machine learning. Five years of physics taught me that the technical work only matters to me when it reaches people.

Open to graduate engineering roles with real client or team contact: application engineering, field service, systems integration and solutions engineering, across Europe, on-site or hybrid. Relocating from September 2026.

English at full professional proficiency · Portuguese native · Russian fluent · German A2

Four cooperative agents controlling the BSM2 wastewater treatment benchmark, trained centrally and executed decentrally. The learned controller reached a lower operating cost than the manual baseline, but it got there by suppressing aeration and undertreating, with ammonia violations 91% of the time. Follow-up experiments with realistic action limits restored compliance, and the controller still did not beat the baseline. The contribution is the evaluation that shows when an apparent cost saving is really just less treatment.

Rigid body kinematics, forward and inverse kinematics, and Jacobian velocity control at about 50 Hz on a real UFactory xArm Lite6 over TCP/IP, with a damped least squares pseudo-inverse for robustness near singularities. MATLAB simulation plus Python on the hardware.

120 runs comparing PPO and DQN across 3 Gymnasium environments, with parallel execution, crash-safe resume and CI. Found a policy collapse failure mode, root-caused it and proposed a test to confirm the explanation.

STL and rolling MAD anomaly detection for stored-grain temperature data. Ramp-gating cut false positives from 125 to 20, which is 84% fewer, at 88% recall and 52% precision. Runs unattended on a small Linux VPS with cron.

Python, PyTorch, Stable-Baselines3, MATLAB and Simulink, pandas, Linux.

I also compete in track and field for Académica de Coimbra.

Reach me on LinkedIn.