Software Consulting & Engineering Requirements to Production

Reproducible ML & distributed storage

2022-2025

← Back to portfolio

  • Distributed storage
  • Erasure coding
  • Reproducible computation
  • ML compilers
  • PyTorch
  • CUDA

Shipped an injectable CUDA library with reproducible PyTorch kernels and built erasure-coded streaming storage for hundreds of gigabytes. Led engineering for decentralized training and helped grow the team from 3 to 25.