Across InspectAI, GitHub Code Chat, Relay AI Support, and Health Programme Analytics, I built machine learning and AI applications spanning computer vision, retrieval-augmented generation, agent tools, and forecasting. InspectAI achieved an image AUROC of 0.9976 on 83 MVTec AD bottle test images, with CPU inference served through FastAPI.
For GitHub Code Chat, I combined BM25 and embedding retrieval with Ollama-generated explanations and evaluated retrieval across five authored files. I also built a typed-tool support workspace with audit traces and tests, and a health analytics pipeline with rolling-origin forecast comparisons.
