Collaborator, LLM/Physics Model Performance

A- Equipment and location

The Contractor will work from their own computer and remotely. The Contractor will comply with Mistral AI’s security policies that will be communicated in writing to Contractor prior to the commencement of Services. Any material changes to such security policies during the Term of the SOW Agreement shall be communicated to the Contractor.

B- Consulting on crash testing efforts

Content

• Collaborate and work closely with Mistral to improve performance of LLMs and/or Physics models for crash testing. This includes but is not limited to:.

◦ Further investigating better model representation for physics models, including tokenization of crash simulation data and mesh-discretization-invariant architectures..

◦ Advising on scaling to larger mesh sizes while adding the domain-specific detail needed for industry-standard fidelity, including more detailed modelling of connectors within the architecture..

◦ Collaborating with the team to understand how solver choices affect simulation outputs, and, depending on findings, on probabilistic methods for capturing output uncertainty..

◦ Advising on future project priorities and evaluation methodology, including industry-relevant metrics and potential use cases unlocked by internal collaborations..

Quality Control, Collaboration & Feedback

• The Contractor should:.

◦ respect the task requirements that will be shared. .

◦ incorporate feedback to iteratively improve the quality and relevance of simulations and/or required outputs..

• Mistral AI commits to answer any pending question and provide feedback, within 2 working days..

C- Roadmap planning for the March 2027 model release

The Contractor will collect necessary information from all stakeholders involved in the March 2027 model release and build a roadmap accordingly alongside the larger team.

D- Involvement in model and data updates

The Contractor may also be tasked with evaluating the capabilities of Mistral AI models on Engineering problems, including but not limited to assessing their performance, accuracy, and robustness. This may involve comparing our models against frontier or baseline models, identifying strengths, weaknesses, and areas for improvement, and providing structured feedback or annotations to support iterative refinement. Based on this feedback, the Contractor will provide advice on new data needs and required model updates.

E- Knowledge dissemination, customer support

• The Contractor will assist Mistral AI’s AI4Eng team in sharing knowledge internally and externally around Mistral AI’s engineering models and products..

• The Contractor will also provide support to Mistral AI’s customers on the deployment and use of Mistral AI’s engineering models and products..