Projects
Updated
My work spans an independent EEG study and language-model implementations. The sleep-stage study’s internal evaluation and the GPT learning project are complete. Publication, external validation and deployment are separate milestones.
Sleep-stage classification — internal study completed
How well does a sleep-stage classifier generalize to unseen people? I implemented spectral baselines, a compact neural model and temporal comparisons, with person-separated development and a reserved final evaluation. Publication and deployment remain pending.
GPT from scratch — learning project completed
A character-level decoder transformer following Andrej Karpathy’s Neural Networks: Zero to Hero. Annotated code and handwritten derivations explain the progression from autograd and makemore to attention and text generation. Read the learning notebook.
Motor-imagery decoding — planned
The direction is to classify imagined movement from EEG. The initial scope, task labels and dataset selection are not fixed. PhysioNet’s EEG Motor Movement/Imagery recordings are a candidate; the existence of a dataset does not establish that a model will decode arbitrary intentions.