The Role
For decades, Cadmould has been one of the fastest and most advanced injection molding simulators on the market, trusted across the plastics industry. Then we built something the field had never seen.
Cadmould AI Solver is the first Large Engineering Model (LEM) for plastic injection molding: a transformer-based neural physics model that delivers high-fidelity results up to 1,000x faster than classical solvers. It turns simulation from a slow validation step into something engineers can explore in real time. It's live as a research preview on our site, and it has already shipped to our first customers.
Teaching a neural network the physics of molten plastic is uncharted territory, and we're still early. You'll work at the core of the model itself: how it's built, how it learns, and the data it learns from. Few teams anywhere are building engineering models like this, and the design space is wide open.
This role can be performed from our office in Würselen near Aachen or remotely from Germany, with occasional travel for team events and on-sites.
What You Will Do
You work alongside our research engineers and domain experts to advance our AI Solver: the model, training, datasets, and evaluation that capture the complexity of real-world injection molding.
- Improve the model. Get to know our transformer-based AI Solver inside out: its architecture, training strategies, data generation, and design trade-offs. Then iterate on architecture, training objectives, and learning strategies to push accuracy and generalization.
- Run and analyze experiments. Design experiments, analyze model behavior, document findings, and translate them into concrete improvements to data, training, and model quality.
- Build shared understanding. Communicate results clearly and contribute to reproducible workflows and shared technical context across the team.
- Strengthen the platform. Extend our training, evaluation, and deployment workflows and infrastructure.
- Pitch in beyond the model. As a small, fast-moving team, we step outside our lane when it counts, from the AI service side to agentic AI features for product experiments. You'll have room to follow the problem wherever it leads.