Apurva MishraThinking through computation

Hi, I’m Apurva.

Better ways to organize computation.

I explore how neural architectures, compilers, and hardware can work together to make AI more capable and efficient.

I like ideas I can build, take apart, and measure.

Let’s compare notes
A schematic GPU floorplanCompute clusters containing arrays of cores surround a shared cache. Memory interfaces sit on either side. This is a conceptual drawing, not a specific GPU. Parallel compute Shared cache / data reuse Compute clustersMemory bandwidthTiled workloads Conceptual floorplan / not to scale
Design the model. Map the work. Understand the machine.
Explore the compute layout

Selected work

Ideas, made executable.

Compilers2024

Rust, sandboxed

Experimental support for running Rust procedural macros in WebAssembly. Built during Google Summer of Code.

GSoC project
Creative engineering2025

typ-browser

What if a browser spoke Typst? A document browser built with SwiftUI and Rust.

Code

Working notes

All writing

Working on ML systems or hardware-aware models?

I’d love to hear what you’re building.