You will decide where we drill next. Each deep drillhole is expensive and reveals a narrow column of rock. You will build planners that search inside our world model to choose the drillholes, surveys and assays that reduce uncertainty fastest within a budget.
What your first year looks like
- Formulate exploration as a POMDP with the applied mathematics team.
- Build planners using MCTS, latent rollouts and model-predictive control.
- Use the world model’s uncertainty to guide exploration, and keep plans robust to model error.
- Evaluate planners against historical exploration campaigns.
- Build ways for geologists to review and override recommendations.
You have:
- A PhD in ML, robotics or operations research, or equivalent experience.
- Depth in model-based RL and planning (MCTS, MuZero, Dreamer, MPC).
- Experience with POMDPs, Bayesian optimisation or active learning.
- Strong PyTorch or JAX.
Nice to have:
- Bayesian experimental design.
- Offline RL or off-policy evaluation.
- Planning in other domains where each query is expensive.
Interested? Write to us at gondwana@altcarbon.com with the role in the subject line.