Featured
My best current work + longform writing
MuBLIS
A framework for instantiating high performance linear algebra (BLAS) libraries, the routines underlying PyTorch and NumPy. Can match OpenBLAS thanks to ARM NEON and AVX2 micro-kernels and compile + runtime specialization.
MinTI-Chess
The world's strongest chess engine for the TI-84 Plus CE graphing calculator, playing at over 2000 Elo. Carefully designed for unique hardware constraints, such as a 24-bit CPU!
Didactic Fast Matmul
7 didactic implementations of matrix multiplication building from naive loops to a parallel kernel that outperforms OpenBLAS.
Work from High School
Earlier projects and writing from before college
Asteroid Impact Prediction (ft. Dr. Biceratops!)
The potentially world-ending YR4 asteroid got a lot of attention recently. How does the math behind predicting asteroid impacts really work?
Algorithmic Gerrymandering
Generating stealthily gerrymandered maps using simulated annealing and voter distribution models.
Polycubes
Why do math when you can write code? Mathematical attempts and code to enumerate polycubes, and computationally finding an error in my homework's answer key.
How does Object Detection with R-CNN work?
An in-depth and intuitive explanation to R-CNN object detection and selective search.