Emanuel Gordis
| Emanuel Gordis | |
| Occupation | Founder of Trim |
|---|---|
| Known for | Co-founding Trim, a foundation model for physics |
| Education | Cornell University (Physics); Princeton University (Physics Research) |
| Alma mater | Cornell University |
Emanuel Gordis is an American entrepreneur, physicist, and software engineer. He is the founder of Trim, a San Francisco-based artificial intelligence company developing a foundation model for physics simulation. Trim was part of the Y Combinator Winter 2025 batch.[1]
Career
Gordis studied physics at Cornell University. He subsequently conducted physics research at Princeton University from 2020 to 2023 and worked in high-energy-density physics at Lawrence Livermore National Laboratory in 2020–2021. Prior to his research positions, he served as a nuclear reactor operator at the Reed Research Reactor from 2018 to 2020. Before founding Trim, he worked as a software engineer at Amazon Web Services (AWS) from 2023 to 2024.
Gordis founded Trim in 2024. The company is building a general intelligence AI model capable of simulating real-world physical systems as they evolve over time. As described by the company, given the starting conditions of a physical system—such as the position of waves on a beach—the model generates how the system progresses forward in time.[2]
Trim's approach addresses computational limitations inherent in traditional physics simulations, which scale exponentially with the number of dimensions and polynomially with grid size. The company's custom architecture, referred to as the Trim Transformer, uses a variant of Galerkin-type linear attention that scales linearly in computation time with respect to both dimensions and grid size. Additionally, whereas traditional simulations require proportionally more time to simulate further into the future, Trim's architecture scales logarithmically with simulation duration. According to the company, these architectural advantages reduce computation time by several orders of magnitude for latency-sensitive applications such as autonomous vehicle path planning and make previously computationally infeasible tasks—such as detecting gravitational waves—more tractable.[2]
One stated application of Trim's technology is in gravitational wave astronomy. Detecting gravitational waves requires comparing observed signals against simulated templates, a process that has been constrained by the computational cost of generating those templates at the necessary frequencies and resolutions.
In July 2025, Y Combinator highlighted that the Trim Transformer reduced memory usage by over 97% compared to a standard PyTorch transformer using softmax attention on two-dimensional Navier–Stokes equations, while also achieving a 6.5-times improvement in time per training epoch.[1]
Gordis is also a co-founder of Nuclear Software, a platform described as a continuous integration system for engineering simulations, co-founded with Javi Vega in 2024.
References