Portfolio

Platform for creating AI simulations. Scaled from zero to 1M+ chat messages with 4K+ MAU and 50m avg. user session time. Achieved $10k revenue; users from Google, Amazon, HubSpot, Stanford, MIT, and hostage negotiators. Built with GPT-4, NextJS, Tailwind, Vercel, and Supabase. Play here!

Scaled to 200,000+ matches played. Co-built with Paradigm an AI 1v1 election game w/ NextJS, Tailwind, and Supabase. See our viral Twitter launch here! We've taken the site down now, unfortunately.

Designed and built an autonomous agent for AI researchers with SWE-Agent, NextJS, Modal, FastAPI, Python, and Supabase. Check it out here.

Piloted an AI investment research platform with an $80B AUM firm. Led design & engineering, mentored on sales and strategy. Website here.

Deep reinforcement learning agent to play Dominion using PPO-Clip and self-play with Python/PyTorch. GitHub here!

Thought-sharing app that sparks interesting conversations with your closest friends. Tested w/ group of ~40 friends before realizing Messenger solves everyone's problems. Built w/ React Native + Firebase. This is the demo and GitHub.

Raised $1.4M to build an NFT marketplace for creators. Built a network of creators with 1 billion+ total followers while operating from the LA Hype House. Led a team of 10 to host celebrity events (e.g. Lil Nas X’s birthday party) and iterate on product with creators.

Wireless AI mushroom chair built with skills in MIT's How to Make Almost Anything: CAD, laser cutting, circuit design, embedded programming, 3D printing/scanning, computer-controlled machining, machine design, & more. Check out the website and GitHub.

Cofounded a collegiate poker platform w/ $60k in funding to 450+ college students across 50+ US colleges. Sponsored by Jane Street, SIG, Flow Traders, and Old Mission Capital.

An iOS app to spontaneously create and explore live events. Built most recent version using Swift and Firebase in 1 month. Had ~130 users at MIT before I realized that this isn't it chief. This is the app on the Apple Store.

World's first Taekwondo accuracy auto-grader. Built using OpenPose, AWS, and Python for MIT's Computer Vision class 6.819. This is the paper on GitHub.