Selected work
A selection of what I've built and researched — from production agent systems to open-source performance work. Deeper case studies are on the way.
Co-author on a filed patent covering the orchestration of multiple AI agents with shared memory and built-in validation — the architecture behind the production agent system I built at NEC.
A retrieval pipeline over a large biomedical document corpus, with a systematic benchmark of embedding models, vector databases and retrieval strategies. Hybrid search produced the largest gain — not a larger model.
M.Sc. thesis: a physics-based simulation that generates training data, then an ML model trained and iteratively optimised on it — covering the full loop from data generation to evaluation.
Implemented a performance-critical octree structure for querying large 3D sensor datasets, cutting query runtime by 50%, validated through systematic benchmarking.