Aman Desai
Research Engineer working at the intersection of ML systems and computer architecture.
CS at UCSB. I work on the systems side of machine learning — performance, kernels, and the architectures underneath.
Research interests
ML systems & performance
What does it take to make a model actually run fast? I'm interested in the gap between mathematical operations and the silicon that executes them — kernels, scheduling, memory hierarchies.
Computer architecture
I want to understand how instruction-level decisions and microarchitectural details propagate up into the workloads we care about. ISAs, branch predictors, caches, the things that make modern CPUs and GPUs feel almost magical.
Compilers & abstractions for accelerators
How do you give a researcher a high-level interface without giving up the performance that motivated using a GPU in the first place? Triton, MLIR, and the ecosystem around them.
Recent writing
All →- Hello, world
A test post that exercises the typesetting — code, math, callouts, and links.
Selected projects
All →- 3D Multi-Species Boid Simulation
Real-time flocking with spatialized audio, written in C++.
- ReSign — Real-time Sign Language Recognition
Webcam-input sign language classifier built at DiamondHacks 2024.
- MIPS ISA Emulator
A small MIPS instruction-set simulator with memory management.