Compute-Aware Hybrid Architecture Search
MIT Deep Learning Research: selectively replace transformer attention layers with linear attention using layer-wise distillation.
- PyTorch
- W&B
- Linear Attention
- Knowledge Distillation
- Research
projects
MIT Deep Learning Research: selectively replace transformer attention layers with linear attention using layer-wise distillation.
CSAIL-affiliated fintech MVP with trade compliance guardrails and portfolio backtesting.
MIT Sandbox-funded e-commerce prototype for custom PCB manufacturing.
Internal admissions tool to distribute applications to reviewers, track scores, and run the admit-list algorithm.
Hidden puzzle hunt on the Splash site. Build and ship puzzles hackers solve for guaranteed admission.
Core HackMIT platform for hacker applications, event management, project submission, and judging.
This website! Portfolio with a terminal mode. Inspired by my friends Sanjith and Arman.
Snake game on an embedded RISC-V processor, implemented in C and RISC-V assembly.
Python Scheme interpreter with tokenize/parse/eval pipeline, frame-based scoping, linked lists, and more with an interactive REPL and command line.
Client/server Star Battle puzzle. ParserLib for .starb files, Express server, and a TypeScript canvas client to place stars and detect wins.