Deep dives into research, design, and engineering -- from neural attention models to brain-computer interfaces.
Developed a Grad-CAM++ based explainability pipeline for chest X-ray classification. Trained a ResNet-50 model on 112k CheXpert images and built saliency alignment metrics that clinicians used to validate model predictions against radiologist annotations.
Key Deliverables
Built GPU-accelerated tensor network simulations for multi-qubit entanglement dynamics. Created a real-time 3D rendering engine in WebGL that visualizes quantum state evolution, Bell inequality violations, and decoherence pathways for up to 8-qubit systems.
Key Deliverables
Designed a brain-computer interface prototype that uses EEG cognitive load signals to dynamically adjust educational content difficulty. Conducted a 40-participant study showing significant retention improvements when content pacing adapts to theta/alpha band ratios.
Key Deliverables
Created an open-source evaluation toolkit for measuring multimodal interaction patterns across ability spectrums. Includes automated WCAG 2.2 auditing, gesture-tracking heuristics, and voice-interaction quality metrics tested across 200+ participants with diverse abilities.
Key Deliverables
Led a discovery-driven redesign of a digital banking onboarding flow. Applied behavioral economics principles (loss aversion framing, progress endowment effect) to reduce drop-off, validated through A/B testing with 12,000 users over 6 weeks.
Key Deliverables
Created a multimodal learning platform designed with accessibility-first principles from the TIILT lab at Northwestern. Integrated adaptive quiz mechanisms, progress dashboards, and collaborative learning spaces informed by learning science research.
Key Deliverables