You'll join the small team building Bombadil and work directly on training: the data the model learns from, the architecture and training decisions that determine whether a run succeeds, and the evaluation work to measure our progress. This is a small team. You’ll drive machine learning projects from idea to production from end-to-end, rather than specializing into one piece of a much larger machine, and you’ll help build systems and practices that scale.
This is a senior, individually-contributing role with no direct reports, reporting to our Head of AI & ML.
Transformer training. Model architecture, data curation, training strategies, and the full lifecycle to production.
Evaluation of the core model. Building and refining evaluations to measure progress of new model capabilities.
Model lifecycle practices. Develop robust, repeatable model lifecycle systems for rapid iteration.
Qualification
Comfortable driving projects end-to-endTechnical or scientific depthExperience with biological
Required
Has built and trained transformer models from scratch, not only fine-tuned them, and can speak concretely to the data, architecture, and training-stability decisions involved.
Roughly 3+ years of hands-on transformer-specific experience is ideal, generally within 5-10 years of overall ML/engineering experience.
Has shipped or maintained ML systems in a practical capacity.
Comfortable driving projects end-to-end, from data collection through evaluation, without a large specialized team around you.
Strong software engineering fundamentals: you build and maintain real training and evaluation infrastructure, not just notebooks.
Technical or scientific depth, not necessarily in CS or ML specifically.
Comfortable being one of a very small number of people responsible for a company's core model.
Experience with biological, environmental, or other scientific data.
This isn't a role for someone who wants a large ML org's infrastructure and specialization already built around them; on a team this size, you'll cover more of the pipeline yourself. It's also not a pure research role: the work has to end up in a model that ships, not just a result that publishes. If you want to bring new skills to a small team building something new, and to build the practice as much as the model, we'd like to talk.
