Research systems, clinical tools, robots, drones, and a fair number of things assembled over a weekend under competition pressure. 19 projects shown.
State-of-the-art text-to-motion generation with 272-dimensional representation, achieving superior animation quality for game engines. Direct SMPL rotation recovery removes IK artefacts from BVH output.
A VQ-VAE producing discrete motion embeddings for transformer-based generation. Codebook resets and EMA stabilisation brought reconstruction error to L2 0.0159 and L1 0.0640.
Converted and optimised a lip-sync generation model for mobile inference with CoreML, running the full pipeline on Apple silicon.
Scans repositories for security issues and surfaces prioritised, explained findings rather than raw rule hits.
An algorithm projecting a 94% improvement in organ-donation success rate while reducing ICU workload. Presented at the Eastern Cardiothoracic Surgical Society.
Chest X-ray classification via fine-tuned ResNet50, reaching 90% accuracy on pneumonia detection, trained on a self-built GPU cluster.
An iOS health monitoring system built around continuous signals and an AI scoring layer, developed at the UNCC AI4Health Center.
A geospatial visualisation of hospital infection data, built to make regional patterns legible at a glance.
Advanced humanoid robotics system integrating AI-powered motion control, computer vision, and real-time sensor fusion, with autonomous navigation and human interaction.
A seventeen-degree-of-freedom expressive head driven by a language model, mapping dialogue state onto facial articulation in real time.
Dell PowerEdge servers and a 40 TB NAS over 10 GbE, used for quantisation and fine-tuning experiments on Tesla P100 GPUs.
A sealed FPV airframe with conformal-coated electronics that survives submersion and keeps flying after it surfaces.
A high-speed digital FPV racing build, tuned end to end from motor selection through PID rates.
Board-level repair and restoration of vintage Apple hardware, including micro-soldering and recapping.
Built Raptor RAG: an advanced retrieval-augmented generation system designed around NASA's space biology data for intelligent ingestion and retrieval. Also selected as a Global Nominee.
Solved complex algorithmic challenges with optimized C++ and Python solutions, demonstrating advanced problem-solving skills.
A cloud-hosted financial tool built and pitched over the course of the Truist immersive programme.
A full-stack build completed with a family team over a single weekend at UNC Chapel Hill.
A Swift iOS app letting users contribute facial data to retrain models on more diverse datasets, with encryption designed in from the start.
Source for many of these projects, plus smaller experiments, lives on GitHub.