Work

Researching and implementing novel State Space Model architectures using Spectral Filtering for efficient long-context language modeling under Prof. Elad Hazan in the Princeton Google DeepMind Lab. Ablating 150M, 500M, and 1B parameter LLMs from scratch.

Increased availability of large 8000+ server distributed system, increasing host redundancy by 15% and eliminating 30% of tickets for on-call engineers.

Built AI solutions for Fortune 500 companies and governments, developing a state-of-the-art Text-to-SQL LLM agent with high accuracy on Spider 1.0 benchmark. Enhanced 3D Gaussian Splatting with Stable Video Diffusion optimization and created an LLM agent for Apple that builds no-code applications with external service integrations.

This was during my spring semester, so I was flying back and forth between San Francisco, UAE, and Princeton (for midterms and finals). I still ended up doing well in all of my classes :)

Worked with Sebastian Thrun and Bichen Wu to incubate a funded consumer video generation startup. Engineered synthetic media generation systems with wav2lip fine-tuning and FLAME 3D facial models, and developed automated video editing algorithms with CLIP-based scene detection deployed on iOS with Metal acceleration.