Daniel Hussey

AI clinical documentation web app for Australian clinicians. It records consultations in-browser, supports on-device transcription via WebGPU and ONNX runtime where available, uses Azure Speech Services as the default cloud transcription path and fallback when local transcription is unavailable, and runs LLM note generation through Azure-hosted services.

Real-time fact-checking PWA for live audio. It transcribes speech, extracts checkable claims, and shows what's supported, disputed, or uncertain. Built with Next.js, Whisper, Grok, and structured outputs.

LLM safety evaluation framework for medicinal chemistry queries. Temporarily private during the NeurIPS '26 submission process. MD research project supervised by Prof. Thomas Balle (USyd).

End-to-end ML demo for classifying ASD vs. control subjects from functional connectivity features in ABIDE rs-fMRI data, with XGBoost, SHAP attribution, and cross-site generalisation analysis.

MRI-only research pipeline for predicting where glioma may recur after surgery. It uses post-operative, pre-radiotherapy T1c and FLAIR MRI, maps later reviewed recurrence labels back into baseline space, and produces voxelwise recurrence-risk heatmaps plus QC reports.

Independent implementation of ACT for learned robotic manipulation policies.