platoseedplatoseed
A foundation model for physics.
Trim is building a general intelligence AI model that can simulate real-world physical systems evolving over time.
Trim is building an AI foundation model designed to simulate real-world physical systems evolving over time. It aims to provide faster, more scalable physics simulations by leveraging a custom transformer architecture.
Trim trains models by running traditional physics simulations and feeding the results into its pipeline. The Trim Transformer uses a custom Galerkin-type attention mechanism, effectively acting as a constant-time lossy lookup table that scales linearly with respect to both dimensions and grid size, enabling lower latency for tasks like autonomous vehicle pathing and complex phenomena such as gravitational waves detection.
Who it’s for: Researchers and engineers who need fast, scalable physics simulations and modeling across domains such as robotics, autonomous systems, and scientific computing.
No explicit funding or hiring mentions in the provided text; references to architecture development and blog explanations implying ongoing R&D activity
Founder of Trim. NRC-licensed RO, published quantum physics with Princeton and astrophysics with LLNL. Cornell Alum.
AI to simulate real-world physical systems
Trim builds an AI model, the Trim Transformer, to simulate real-world physical systems evolving over time. It claims linear-attention enables faster, scalable simulations across high-dimensional grids, with applications including autonomous navigation, molecular modeling, climate modeling, financial modeling, fusion plasmas, and gravitational waves; the launch highlights improved memory efficiency and speed over standard transformers and invites users to explore on GitHub and their blog.
Formerly “Nuclear Software Solutions”, “Nuclear Software” · why startups rename →
Physics-based foundation models for CAD optimization.
Specialized AI for Every Job
