Emanuel Gordis

Emanuel Gordis

Emanuel Gordis

San Francisco, CA


About

I grew up in Berkeley, CA and was homeschooled. I build models that learn to simulate complex physical systems faster than classical methods allow.

Most recently, I founded Trim (YC W25), where I studied how different sequence-modeling approaches behave on physical systems. I developed a transformer-based physics world model using a custom attention mechanism, resulting in a 97% reduction in memory usage and a 6.5Γ— speedup on Navier-Stokes benchmarks. The trim-transformer package is open source.

Before founding Trim, I conducted astrophysics research at Lawrence Livermore National Laboratory β€” simulating thermonuclear detonation in Type Ia supernovae β€” and quantum mechanics research at Princeton with Prof. Michael Littman, studying Rydberg atom quantum states. Both projects resulted in publications. I also built computer vision and ML infra at AWS.

I studied physics at Cornell and spent a formative stretch at Reed College, where I became the youngest federally licensed reactor operator in the United States.


Publications

Stability and Convergence of Nuclear Detonations in White Dwarf Collisions P. Anninos, D. Cruz-Lopez, B. Jiang, E. Gordis Β· arXiv, 2025

The hydrogen Rydberg atom viewed through the lens of the old quantum theory M. G. Littman, E. Gordis, P. Zhelnin, J. Arnold Β· American Journal of Physics, 2023


Research

Founder β€” Trim (YC W25) Β· 2025–2026

Physics world model with a custom attention architecture. Achieved 97% memory reduction and a 6.5Γ— speedup on Navier-Stokes benchmarks over standard transformers. Open-sourced as trim-transformer, a drop-in replacement for standard PyTorch transformers.

Unified ANIE, backpropagation through time, and video transformers as a 2Γ—2Γ—3 design grid, ablated all 12 configurations across PDE benchmarks including Navier-Stokes, Burgers, and Shallow-Water. Found that distributional shift dominates performance. Shared weights with no block residuals were consistently near-optimal, suggesting an inductive bias toward iterative refinement.

Quantum Mechanics Researcher β€” Princeton University Β· 2020–2023

Investigated Rydberg atom quantum states using Bohr–Sommerfeld semi-classical analysis, with Prof. Michael Littman. Published in American Journal of Physics.

High-Energy Physics Researcher β€” Lawrence Livermore National Laboratory Β· 2020–2021

Simulated Type Ia supernovae white dwarf collisions, modeling magnetic field effects on thermonuclear detonation propagation. Published in The Astrophysical Journal.


Other Experience

Software Engineer β€” Amazon Web Services Β· 2023–2024 Built load balancing infrastructure for ML training queues and led a live production database migration from SQL to DynamoDB with zero downtime.

Reactor Operator β€” Reed Research Reactor Β· 2019–2020 Youngest federally licensed reactor operator in the United States.


Education

Cornell University β€” B.A. Physics, Concentration in Computer Science Β· 2022