Interested in the foundations and applications of ML — from the mathematical principles that explain why models work to building systems that work in production
-
TEMPO: Regime-Aware Operator Learning with Locally Adapted POD Bases
(in progress) -
Diffractive Deep Neural Network (D2NN)
All-optical inference via Rayleigh–Sommerfeld wave propagation. Physics-informed phase-mask initialisation using the Gerchberg–Saxton algorithm.
Winner — General Project Competition -
Spin Glasses & Neural Networks
Multilayer Sherrington–Kirkpatrick model as a theoretical framework for deep networks. Connects replica symmetry breaking with loss landscape geometry.
Best Talk — 67th Scientific Conference -
Protein Folding
Coarse-grained protein simulation via Monte Carlo sampling (Metropolis algorithm) with Lennard-Jones and harmonic potentials. Analysed thermodynamic observables including heat capacity and equilibrium bond geometry.
Winner — General Project Competition -
LLM Infrastructure (Industry)
Secure local inference stack (vLLM + Ollama). Multi-agent document analysis pipeline. MLOps with Kubernetes orchestration.
PyTorch · NumPy · HuggingFace · vLLM · Ollama · Kubernetes · Docker · WandB