Projects

Automated scoring of a standard neuropsychological test by building and comparing deep neural networks on an enhanced figure dataset, then applied interpretability methods to make the normally black-box models explainable.

Forecast inflation with ARIMA/SARIMAX and LSTM models, recommending LSTM for short-term and actuarial models for long-term forecasts; derived engineering-insurance premium pricing for two case studies using a hybrid regression model combining weather and inflation data.

Designed unsupervised NLP models for intelligent notification handling on mobile devices — TF-IDF feature extraction and short-text topic modeling (GSDMM, GPU-DMM) benchmarked against LDA.

60% improvement in clustering performance over the LDA baseline

Improved the vision, AI, and control stack of a robotic billiards player, identifying manual bottlenecks and re-implementing weak points to increase stability and performance.

Contributed to one of the first autonomous robots for inspecting oil and gas plants, built for Total's ARGOS Challenge; continued onto a second-generation robot with Mitsubishi Heavy Industries, improving its design and control.

Proposed a theoretical model predicting contact force between a multi-DOF tracked vehicle and an obstacle across vehicle sizes and motions, validated against physical experiments with a real tracked vehicle.