Marco Cusumano-towner
Research Scientist at Databricks
United States
Summary
🤩
Rockstar🎓
Top SchoolMarco Cusumano-towner is a research scientist with 11 years of experience bridging probabilistic programming, reinforcement learning, and autonomous systems, currently at Databricks after research roles at Apple and MIT. He earned a PhD from MIT and led the creation of Gen, a widely used probabilistic programming system (≈1.8k stars) where he implemented core generator infrastructure and novel backprop primitives. His recent work includes long-horizon RL for interactive LLM agents and large-scale multi-agent self-play for robust autonomy (ICML 2025), reflecting a focus on scalable learning for complex decision-making. Comfortable moving between theory and production, he has applied ML to genomics, robotics, and driving, and often blends program synthesis and information-theoretic ideas into practical systems. An understated strength is his track record of shipping foundational backend primitives (parsers, IRs, and inference hooks) that enable higher-level research to scale.
11 years of coding experience
13 years of employment as a software developer
B.S. Electrical Engineering and Computer Science, B.S. Electrical Engineering and Computer Science at University of California, BerkeleyDoctor of Philosophy (Ph.D.) Electrical Engineering and Computer Science (AI), Doctor of Philosophy (Ph.D.) Electrical Engineering and Computer Science (AI) at Massachusetts Institute of TechnologyM.S. Computer Science, M.S. Computer Science at Stanford University
Github Skills (3)
probabilistic-programming10
python3
machine-learning3
Programming languages (5)
Github contributions (5)
probcomp/Gen.jl
Aug 2018 - Feb 2022
A general-purpose probabilistic programming system with programmable inference
Role in this project:
Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:13 releases, 24 reviews, 978 commits in 3 years 6 months
Contributions summary:Marco made several contributions to the probabilistic programming system. They implemented core components related to basic generators, including the foundational infrastructure for parsing abstract syntax trees into intermediate representations. The user also made changes related to incorporating the idea of *change* and implementing *backprop* for these primitives.
roboticsdifferentiable-programmingjulia-languagedeep-learningsystem-programming
probcomp/GenExamples.jl
Feb 2021 - May 2021
Gen examples with a Travis CI build that tests that they run
Contributions:32 commits, 4 PRs, 20 pushes in 3 months
testingtraviscontinuous-integrationtravis-cigen
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.