About the job
Build Models Where None Exist Yet. At Flagship Pioneering, we create companies from first principles. Within Flagship Labs, small founding teams define new technical theses, test them rapidly, and build ventures around breakthrough ideas. We are forming a machine learning team inside a newly launched venture, Flagship Labs 120. Our work focuses on extracting latent structure from information-rich measurements of complex physical systems—often requiring mechanism-informed modeling, thoughtful inductive bias design, and principled approaches to inverse problems. This is a zero-to-one role focused on modeling innovation rather than routine optimization. You’ll design, prototype, test, and refine new approaches that help define the technical foundation of a platform from day one.
Responsibilities
Develop and iterate on ML models for complex measurement data, from representation design through validation
Design objectives and architectures that respect known constraints, symmetries, or latent structure in the data
Explore and compare modeling strategies, balancing strong baselines with more experimental approaches when appropriate
Investigate model behavior and failure modes to improve robustness and interpretability
Collaborate closely with experimental and technical teammates to align modeling with data generation
Contribute to shaping the long-term ML strategy and technical direction of a new venture
Qualifications
Minimum
Strong hands-on experience building and training modern ML models
Fluency in Python and at least one major ML framework (e.g., PyTorch or equivalent)
Experience working with real-world or experimentally generated data
Ability to design, run, and interpret ML experiments
Comfort working in practical development environments (e.g., cloud infrastructure, experiment tracking, reproducible workflows)
Preferred
Experience with inverse problems, latent-variable inference, or structured generative modeling (e.g., diffusion or flow-based methods)
Familiarity with geometric or symmetry-aware architectures
Experience incorporating physical or structural constraints into learning systems
Experience working with time-series or high-dimensional signal data
Exposure to biology, chemistry, physics, or related sciences