An automated pipeline to translate videos into target languages while maintaining precise lip-syncing and speaker voice identity.

• Created syllable-matching duration-alignment using NLP synonym swapping & beam search.
• Implemented dynamic lip-syncing using audio-conditioned U-Nets to manipulate VAE latent spaces.
• Integrated zero-shot voice cloning using acoustic encoders.

Fine-tuned a Large Language Model to output logical and legal chess movements using reinforcement learning.

• Utilized Group Relative Policy Optimization (GRPO) and LoRA for resource-efficient alignment.
• Increased correct move generation by 95% over base models.
• Achieved reasoning capabilities through strategic positioning.

Wearable computer vision glasses aiding visually impaired individuals to navigate physical spaces safely.

• Engineered object detection & obstacle mapping using YOLO and OpenCV.
• Implemented audio translations using ElevenLabs and OpenAI Whisper.
• Built mobile streaming client in React Native to forward high-FPS camera streams.

A full-stack multiplayer civics education web application scaled to over 100,000+ active users.

• Built synchronous multiplayer gameplay using WebSockets/Socket.io.
• Engineered React front-end, Node.js game controller, and Python Flask API.
• Presented platform on NPR (Here & Now) to millions of listeners.

Computer vision lane detection and PID navigation systems for autonomous vehicles in the CARLA simulator.

• Designed robust lane detection algorithms in OpenCV to handle complex lighting variations.
• Developed adaptive PID control loops for vehicle stabilization and racing-line navigation.
• Placed third globally and published research highlights on the Berkeley ROAR site.