I'm Eymene Khiat.

A multi-agent reinforcement learning framework (MAPPO) that lets warehouse robots plan their own routes. Bootstrapping with behavioral cloning and DAgger from a centralized planner gave 30% faster convergence and 72% better deadlock handling, validated through ablation studies on a Dec-POMDP formulation.

An end-to-end execution system that reads biometric readiness (HRV, sleep) and real behavior, then adjusts planning intensity to match. Built on robust statistics (median, MAD, tanh normalization) with a glass-box policy engine governing goal capture, roadmaps, and pacing.

A high-throughput ML pipeline that models antibiotic resistance from clinical and bioinformatics datasets. SHAP makes the predictions interpretable for medical use, and careful preprocessing and outlier detection quantify how data quality affects model performance.

Non-linear state estimation with Extended and Unscented Kalman Filters to localise a robot in noisy environments, paired with reinforcement learning for adaptive motion planning over multi-threaded sensor fusion.

A conversational analytics system that lets non-technical users explore business data in plain English. Questions go in, queries and visual answers come out.

Real-time object detection and counting inside custom polygonal zones drawn over video feeds. Useful for footfall counting, safety zones, and traffic analysis.