Background

About me

I'm a software engineer at YouTube, currently working on tools and infrastructure for playback. Previously, I developed quantization frameworks for YouTube's hardware accelerator team and built an open source conversational AI accelerator at Microsoft CoreAI.

I studied EECS at UC Berkeley, where I explored the intersection of machine learning and visual systems through research in autonomous racing and theoretical neuroscience. I'm also passionate about educational outreach to younger students, and started a free machine learning bootcamp for high school students.

Outside of engineering, I care a lot about design and aesthetics and want to continue building my intuition for well-designed solutions. I also love running, exploring the outdoors, and curating Spotify playlists!

Machine Learning  ·  Generative Systems

N3CTAR — Neural Cellular Automata

An interactive simulation of neural cellular automata: neural networks learn local update rules that, when applied to a grid of cells, produce complex self-organizing patterns from a single seed state. Explores emergent behaviour at the boundary of ML and generative systems.

View live demo

Azure  ·  NLP  ·  Conversational AI

Azure Language Conversational Agent

An accelerator for building enterprise conversational agents powered by Azure Language Services and OpenAI. Combines intent recognition, entity extraction, and generative AI to enable natural, context-aware dialogue at scale.

View on GitHub

Research  ·  Computer Vision  ·  Representation Learning

Compositional Factorization of Visual Scenes

A system for visual scene parsing that encodes sparse latent features into high-dimensional vectors, then factorizes them to recover scene content. Integrates convolutional sparse coding with resonator networks to increase representational capacity and reduce collisions in combinatorial search.

Read on arXiv
Aug – Oct 2025

Software Engineering Intern — YouTube Hardware Acceleration

Google

Built a custom IR parser to extract JAX model topologies and convert them into silicon protobuf format for Argos, YouTube's custom video transcoding ASIC. Designed and benchmarked PTQ and QAT pipelines for JAX models, mapping QKeras quantization configs to AQT with bitwise weight accuracy.

May – Aug 2025

Software Engineering Intern — AI Language Team

Microsoft

Integrated a multi-agent conversational framework via Semantic Kernel with Azure AI Translation and Language Services into an official AI Foundry template. The contribution doubled GitHub stars and forks from 30 to 75+ within three months.

May – Aug 2024

Machine Learning Engineering Intern

Autodesk

Built a GPT-4 Wiki security screening pipeline that processed 15,000+ pages to surface confidential data. Finetuned DeBERTa and Llama3 on AWS Lambda to auto-route support tickets, reducing engineer overhead by 80% per ticket.

Mar – Dec 2024

Perception Team Researcher

AI Racing Tech — Top US Indy Autonomous Racing Team

Implemented a YOLO-based annotation pipeline to label opponent Formula 1 cars across 40TB of race data. Optimized bounding box sizing and confidence thresholds for real-time opponent vehicle detection.

2022 — 2026

B.S. Electrical Engineering & Computer Science

University of California, Berkeley  ·