Project Manager, Silent Speech Decoding

Led development of multimodal silent speech decoding pipeline by synchronizing kinematic & visual signals. Explored cross-modal contrastive learning to enable robust classification under noisy, partially observed sensor data.

Designed Hybrid LinUCB contextual bandit to adaptively select autonomy levels for Google Calendar AI agent, modeling user trust as a time-evolving state from real-time interaction feedback.

Engineered trust-aware reward functions and feature pipelines integrating user history, task context, and uncertainty-driven exploration to safely adapt agent behavior and maximize long-term user acceptance.

US Utility Patent 63/614,530, California State Seal of Civic Engagement & Explorer Award Sponsored by Google. Device to detect overhead obstacles for the blind; CV, Arduino Uno & Nano, ultrasonic sensors, & vibration motors.

Built 4 multimodal perception systems for assistive navigation in audio, visual, and language domains.

Implemented Discrete Fourier Transform-based feature extraction, sequential neural networks, NLP & transformers.

Co-authored binational AI and climate-technology policy between the Government of NSW, Australia and the State of California, informing data-driven frameworks for climate resilience, energy systems, and technology deployment.

Defined collaboration on ML-enabled grid optimization, emissions quantification models, & ESG risk assessment.