Jacob Marshall boat

I like building models/software/companies

I've spent time working across the AI research spectrum, from training and optimizing models to building agentic systems. My main research interests are in reinforcement learning and computer vision.

I am a co-founder of planweave, we're focused on building agentic systems for automating building code compliance.

previously:
  • employee #7 at sieve where I led the technical effort to scale our video datasets offering to leading frontier labs from 0 to $X0M in revenue. also trained NeRF models for cheap lip-syncing
  • quant-dev / swe at xr trading
  • dropped out of a PhD program to start a quant trading startup
open-source work:
  • turbozero ★ ... - GPU-accelerated AlphaZero implementation in JAX. cited by [1]
  • mctx-az ★ ... - Monte Carlo Tree Search in JAX with features for effective AlphaZero
  • turbo-neat ★ ... - GPU-accelerated 'NeuroEvolution of Augmenting Topologies' implementation in JAX.

other interests:
  • intelligence
  • poker
  • aquariums and aquascaping