Aayush Gupta

Right now I advise Extrapolate, which builds RL training environments for long-horizon AI research. In the past: evals and verifiers for AI-driven hardware design at Normal (we made FrontierCAD, new part and assembly verifiers for CAD generation), computer use agents and environments at Proteus, ultrasound imaging and neuromodulation at aleph, perceptual hashing (adversarially robust image provenance), open education projects, open source zk cryptography, multi-tissue aging research, equivariant molecular neural networks, applied ML, and healthcare. I also like blogging about conviction and movie/travel recs. I also deeply care about open education for MIT and Harvard, and give my money away as grants for cool open source work.

Best Posts

  • Blog
    Writing about AI evals, conviction, media, and obsessions.
  • FrontierCAD
    New verifiers to test whether frontier models can build CAD parts and assemblies from drawings.
  • Thoughts for Curious Rabbit-holers
    A live post of short updates across topics.
  • Conviction
    On building and testing conviction in ideas.
  • Movie/Travel Recommendations
    Novelty-focused recommendations.

Projects