Your Impact at LILA

Lila Sciences builds AI systems that accelerate discovery across the physical and life sciences. Within Physical Sciences AI, we work with experimental teams to build autonomous discovery agents that decide which experiments Lila's AI Science Factories should run next.

Real scientific campaigns rarely look like textbook optimization problems. A single campaign can mix large discrete choices (which components, which recipe) with continuous settings (amounts, temperatures, times), pursue several objectives under hard constraints, and draw on evidence from literature and characterization data. Every experiment costs real time and materials. Off-the-shelf methods often struggle here, and they ignore the domain structure that could guide the search.

As a Bayesian Methods Co-Op, you will work alongside research scientists and engineers on a focused project within one broad question: how AI systems can make better experimental decisions. Depending on your background, that could mean building agents that choose the next experiment, testing when in-context learners can be trusted as decision-makers, or developing Bayesian strategies for tight experimental budgets. Your methods will be tested against live campaigns.

What You'll Be Building

• Design and benchmark experiment-selection strategies that balance exploration and exploitation under realistic budgets
• Build autonomous discovery agents that propose the next experiment from surrogate models
• Develop in-context learning approaches inspired by Bayesian inference, evaluating calibration and robustness
• Run retrospective and live evaluations on active campaign data with clear baselines and metrics
• Present findings to ML and experimental teams, with the option to contribute to a publication

What You'll Need to Succeed

• Pursuing a Master's or PhD in statistics, ML, CS, operations research, applied math, or a quantitative science
• Solid grounding in Bayesian inference and decision making, including priors and posterior predictive distributions
• Research experience in Gaussian processes, Bayesian optimization, bandits, reinforcement learning, or probabilistic modeling
• Strong Python skills with PyTorch and probabilistic libraries such as Pyro, PyMC, or BoTorch
• Comfort working with small, noisy, and heterogeneous datasets
• Clear communication with both ML and experimental collaborators

Bonus Points For

• Applying these methods to real problems in materials science, chemistry, or another physical science
• Familiarity with in-context learning as implicit Bayesian inference, meta-learning, or prior-data fitted networks
• Optimization over categorical or combinatorial spaces, including mixed-variable, multi-objective, or batch settings
• Reinforcement learning for sequential decision making or Bayesian experimental design
• Training models on multimodal scientific data such as text, images, spectra, or simulations