Mahdi Ahmadi

Built a reproducible Python/Slurm pipeline for Lumerical FDTD data: 2,003 frequency samples at fixed geometry and over 70,000 samples across geometry sweeps. Compared magnitude- and phase-aware representations, dimensionality reduction and clustering. Developed a multivariate shared-pole model, selected complexity through blocked five-fold cross-validation and tested stability through deletion refits. Eight resonance families were consistently supported; a ninth remained a weak candidate.

At the BTU AI Lab, in cooperation with Lebenshilfe e.V., developed a YOLO-based computer-vision system for real-time detection and counting of screws in assembly trays. Prepared and annotated the image dataset, automated sub-tray extraction, and evaluated detection and counting robustness under occlusion and reflections using reproducible experiments.

Built reproducible PyTorch experiments on a quantized BitNet-style language model under simulated RRAM-inspired weight drift. Perturbed ternary weights after quantization and before 2-bit packing, controlling layer, drift level, direction and random seed. Stress-tested a matrix of approximately 17.7 million parameters; 5% drift modified 884,736 weights. Observed gradual degradation: text generation continued at 80-90% drift, with more repetition and reduced coherence.

Developed modular Python tools for Raman spectroscopy and diffraction data, including baseline correction, cosmic-spike removal, peak detection and fitting. Used dimensionality reduction and spatial clustering to analyse Raman maps and inspect unusual spectral patterns and measurement artefacts. Built a PyQt interface for visual inspection and iterative analysis with materials researchers.

Completed September 2026; thesis defended on 24 September 2026 and graded 1.1. Thesis: Identification and Classification of Photonic Modes in Complex Nanostructures Using Machine Learning. Developed a Python/HPC pipeline and multivariate shared-pole model for complex electromagnetic fields, with blocked cross-validation and stability analysis. Degree certificate in process; official transcript available.