things i'm trying to understand.

Replicated the problem at scale on 83 subjects (PhysioNet Motor Imagery), then proposed DA-EEGNet — a domain-adversarial EEGNet using gradient reversal to learn session-invariant identity representations. It reached 85.11% leave-one-session-out accuracy against 71.53% (SVM) and 63.95% (EEGNet) baselines. A first systematic cross-session channel ablation showed 8 electrodes retain 80% of full 64-channel performance, with a sharp cliff below 4.

Characterized subject-level variability in EEG under identical preprocessing and found that inter-individual differences were signal, not noise. Presented the findings at a national neuroscience conference.

Engineered personalized, neuroscience-informed learning tools and grew a 4,000+ parent community around the platform.

Researched object-recognition pipelines for smart eyewear targeting industrial safety applications, and contributed to assistive-technology projects for visually impaired users.

Integrated object recognition into an AR learning tool at the New York Academy of Sciences. Selected as a Finalist and International Runner-Up.