Grafton Sciences is building physical superintelligence through autonomous laboratories that learn through interaction, starting with disease and energy. The company aims to build the first system capable of true scientific discovery.
Requirements
- Fluency in PyTorch or JAX and production-grade Python engineering.
- Familiarity with Bayesian optimization, active learning, and uncertainty quantification.
- Experience deploying models that interact with real or simulated hardware.
- Work in agentic AI, robotics autonomy, computational chemistry, or materials ML.
- Contributions to open-source RL, optimization, or simulation frameworks are valued.
Responsibilities
- Design and train RL and optimization algorithms that guide physical experiments.
- Develop surrogate models and digital twins to accelerate experiment planning and control.
- Implement LoRA/QLoRA fine-tuning pipelines and model-evaluation harnesses.
- Integrate LLMs with retrieval and agentic frameworks (LangGraph or similar).
- Collaborate with software and robotics teams to close the loop between digital reasoning and physical actuation.
- Build systems that are transparent, auditable, and robust under real-world uncertainty.
Other
- MS/PhD in Computer Science, Applied Physics, or related discipline.
- 3 to 7+ years in applied ML, reinforcement learning, or scientific machine learning.
- A temperament suited to early-stage environments: relentless follow-through, intellectual curiosity, and calm in ambiguity.
- Candidates must demonstrate clear evidence of systems-level thinking and executional excellence; formal technical degrees do not impact evaluation.
- We’re looking for people who thrive in demanding environments — researchers and engineers who could work anywhere but want to work on something that actually changes the pace of scientific discovery.
