MLFF Distillation & GCMC Integration

About the role

You will own the end-to-end development and deployment of machine learning force fields specifically designed for high-throughput Monte Carlo simulations of gas adsorption in porous materials. You will distill large, complex equivariant models into lightweight student potentials that retain physical fidelity while meeting severe computational constraints. Your daily work will focus on integrating these distilled models into the kUPS simulation framework and ensuring they are robust enough for production use. You will own the validation strategy that proves the distilled MLFFs are superior to classical force fields for guest-host interactions in metal-organic frameworks. You will collaborate directly with computational chemists to design reference datasets and benchmark protocols that the wider scientific community can reuse. You will own the performance profiling work that identifies bottlenecks in the Monte Carlo inner loop and guides optimization efforts. Your contributions will culminate in a foundational capability that enables scalable, accurate screening of MOFs for gas storage and separation applications. You will be an engineering intern within the chemistry team at CuspAI, working alongside some of the most cited researchers in AI and materials science.

Key facts

What you'll do

• Distill MLFFs into student potentials that are explicitly optimized for speed and stability in Monte Carlo sampling.
• Curate, version, and document training and validation datasets with rigorous tracking of the distillation protocol and any active-learning loops used to close coverage gaps.
• Run head-to-head validation campaigns that compare the distilled MLFF against classical force-field baselines across a curated set of guest molecules, carefully characterising accuracy, throughput, and failure modes.
• Profile and optimise the end-to-end simulation pipeline, with particular attention to the MC inner loop where MLFF inference cost dominates resource usage and limits scalability.
• Benchmark accuracy and speed trade-offs systematically, documenting clearly where the distilled model fails and under which conditions those failures occur.
• Collaborate with computational chemists on reference data generation, benchmark system selection, and validation strategy to ensure scientific rigor and reproducibility.
• Contribute to a publication that establishes MLFF-driven GCMC as a credible and scalable approach for MOF screening.
• Partner with infrastructure engineers to ensure that the distilled potentials integrate cleanly into kUPS and support high-throughput campaign execution.
• Analyse failure modes in distillation and validation, proposing targeted improvements to training data, model architecture, or loss functions.
• Maintain detailed documentation of methods, parameters, and results so that findings are transparent and reusable by other teams and external collaborators.
• Translate scientific requirements from chemistry and materials science into concrete engineering tasks for simulation and ML workflows.
• Support the creation of reusable components and tooling that enable other teams to adopt MLFFs in their own projects without repeated custom development.
• Track and report on key performance indicators such as simulation throughput, wall-clock time per GCMC cycle, and accuracy against experimental isotherms.
• Act as the primary point of contact for questions and clarifications related to MLFF distillation, integration, and validation within the simulation stack.

Requirements

• Currently enrolled in (or recently completed) a PhD or Master's programme in a relevant quantitative field (Physics, Chemistry, Chemical Engineering, Computational Science, Machine Learning, or similar).
• Experience in adsorption modelling at the atomic scale, including the physical interpretation of adsorption isotherms and uptake metrics.
• Hands-on experience with molecular simulation methods, including GCMC, MD, or both, and a clear understanding of their respective strengths and limitations.
• Comfortable working in Linux environments and managing simulation campaigns at scale, including job scheduling, data organisation, and reproducibility practices.
• A genuine interest in the application of machine learning to chemistry and materials science, with a desire to see rigorous scientific validation accompany any model deployment.
• Strong programming skills, preferably in Python and/or C++, with the ability to read, modify, and debug code that interfaces with simulation frameworks.
• Excellent written and verbal communication skills, with the ability to explain technical trade-offs to both technical and non-technical stakeholders.
• A disciplined approach to version control, testing, and documentation, ensuring that code and data are well organised and easily auditable.
• Willingness to work closely with cross-functional teams, adhering to agreed timelines and contributing constructively to collaborative problem-solving.

Nice to have

• Familiarity with modern MLFFs, including recent architectures and training strategies from the literature.
• Experience with knowledge distillation or other model compression techniques for scientific machine learning.
• Experience with active learning workflows for atomistic data, including strategies for uncertainty quantification and targeted data generation.
• Familiarity with density functional theory data generation and the practicalities of curating atomistic datasets from electronic structure calculations.
• Direct experience with established simulation packages such as LAMMPS, RASPA, or similar, and an understanding of their input/output conventions.
• Background in gas adsorption, MOFs, or porous materials, including knowledge of common structure metrics and characterization methods.
• Familiarity with classical force fields used in MOF simulation and an understanding of how parameters influence predicted adsorption behavior.
• A track record of published research at top-tier ML or computational chemistry venues, demonstrating an ability to connect theoretical concepts to practical implementations.

Practical notes

• This is a 3-month internship position.
• You would be joining as an engineering intern within the chemistry team at CuspAI.

What we offer

• A competitive

salary: We value and reward impact and growth.

• Equity in CuspAI: You have a stake in the success of the company.
• Time off to stay fresh: 28 days holiday (DE, NL, UK) or 21 days holiday (JP, SG, US), in addition to local public holidays.
• 'Gold Standard' parental leave: 26 weeks (primary caregiver) and 12 weeks (secondary caregiver) at full pay - we look after you and your family while we work on the most important materials discovery problems together.
• Professional development budget: We invest in your career development so you can stay up to date with the latest industry knowledge or add to your skills to increase impact and growth.
• Solve meaningful problems: See how your work has a direct impact on advancing materials science and solving sustainability and climate-related problems through the creation and application of bleeding-edge SOTA technology and frameworks.