Research Engineer, Scientific Agents & Simulation

The role

A scientist takes an ambiguous objective, decomposes it into tractable questions, chooses methods, writes code, runs experiments, interprets the results, diagnoses failures, and decides what to try next. Your job is to make that loop executable by machines.

Our current work focuses on quantum chemistry and molecular simulation. The larger goal is to build systems capable of performing increasingly large portions of computational scientific research. We are looking for an engineer to help us build that system.

You will build the infrastructure connecting frontier AI models to scientific software, simulators, code execution, datasets, evaluators, and large scale compute.

The goal is not an AI that knows science or can perform literature retrieval. It is an AI system that can do scientific work.

You will work directly with the founders and scientists and have substantial ownership over the architecture of our scientific agent system.

What you will do

• Build agents that turn open ended scientific objectives into computational research workflows.
• Decompose difficult scientific problems into smaller computationally verifiable subproblems.
• Build systems for launching experiments, inspecting results, diagnosing failures, and deciding what to run next.
• Integrate our electronic structure models and other simulators as tools and verifiers.
• Build evaluations that measure whether agents actually solve scientific problems.
• Build robust infrastructure for long-running jobs across GPUs and CPUs, including checkpointing, provenance, reproducibility, and failure recovery.
• Work directly with scientists to turn expert reasoning into tools, workflows, and increasingly autonomous research systems.

What we are looking for

You should be excellent at Python and comfortable owning systems end-to-end.

Strong candidates will have experience with the following:

• Building LLM agents, tool use, or coding agents with frameworks such as DSPy
• Building, training, and evaluating models in JAX or PyTorch
• Extending neural network architectures such as transformers and message passing graph neural networks
• Post-training LLMs with reinforcement learning techniques such as RLVR

Experience in computational chemistry, physics, biology, materials science, or another computational science is valuable but not required.

You may be a particularly good fit if you have built agents that interact with real tools, infrastructure for simulations or ML research, or autonomous systems that operate for hours rather than seconds.