Open-source simulation tools have become essential in science and engineering. They offer flexibility, transparency, and the ability to tackle complex problems at multiple scales. However, using them efficiently often requires deep technical knowledge, careful input preparation, and a good deal of trial and error. This learning curve can discourage researchers who might otherwise benefit from their power.
One of my current interests is exploring how Large Language Models (LLMs) can be used to reduce this barrier. By creating natural language interfaces that can understand, modify, and generate input files or scripts, LLMs can help users interact with simulation tools more intuitively. Whether it’s defining boundary conditions, checking mesh settings, or translating a conceptual model into code, these systems have the potential to make advanced tools more accessible to a broader community.
In parallel, I am also investigating the use of Physics-Informed Neural Networks (PINNs) to support and accelerate deterministic simulations. By embedding governing equations directly into the learning process, PINNs offer a way to solve or approximate complex systems without relying entirely on mesh-based solvers. This approach can help improve performance, reduce computation time, and even extend simulations into regimes that are difficult to reach using classical methods alone.
Together, these directions reflect a broader interest in combining AI with scientific computing. The goal is not to replace established tools, but to enhance and support them, making it easier to explore ideas, run simulations, and extract insights in a way that is both efficient and user-friendly.
If you are working in this area or curious about building bridges between data-driven models and physically grounded simulations, feel free to reach out. I believe this is a field where collaboration and openness will play a key role.
Add comment