Senior ML Research Engineer

• You'll own real equity in what you build. We offer meaningful stock options because we believe the people building this company should share in what it becomes. We want teammates who think like owners, and we structure compensation to reflect that.
• Outstanding benefits. Private medical and dental with Bupa from day one, paid by Outpost, and dependents covered at 50%. Pension with a 3% employer contribution. Life insurance, income protection and critical illness cover, all employer paid. Cycle to Work scheme. 25 days holiday plus bank holidays, your birthday off, and a paid winter break between Christmas Eve and New Year.
• An ML Lab-in-the-Loop. Your work feeds directly into Outpost's AI platform, and the platform feeds back into the next experiment. The loop between the wet lab, the data and the models runs in days, not years, and you'll iterate inside it whichever side you sit on.
• How we work. Expected in the office Monday, Wednesday and Friday, with flexible hours around a 10am to 4pm core and up to two weeks a year working from anywhere. Home office stipend and company computer. We cover conference costs and want you presenting your work, and every quarter you get a budget for drinks or coffee to learn from peers at other companies.
• Small team, outsized reach. You're joining a small founding team backed by top-tier investors with deep connections across AI and bio. The science you do here will directly shape how pharma and consumer companies understand molecule and microbiome interactions.

Biological data is noisy, high-dimensional, and shaped by the experimental process as much as by biology itself. We are hiring a Senior ML Research Engineer to develop the models and research infrastructure that turn our multi-omic datasets, including metagenomics, metabolomics and 16S, into reliable scientific insight.

You will own high-impact research questions from hypothesis through publication-quality evaluation, while building reproducible training and inference systems that let the team move quickly and trust its results. This role is central to translating our proprietary data and experimental capabilities into a durable scientific and product advantage.

Here is our timeline for hiring this role:

• Design and train novel machine-learning models for messy multi-omic data, including metagenomics, metabolomics and 16S, grounded in a strong understanding of biological systems.
• Own research questions end to end, from hypothesis and dataset design through training, evaluation and write-up.
• Set rigorous evaluation standards, accounting for batch effects, leakage, compositional data and correlated samples, to distinguish real results from artifacts.
• Turn research into reproducible infrastructure through versioned training and inference pipelines, extensible codebases and robust experiment tracking.
• Work closely across data and lab teams to shape experiments, define data handoffs, and communicate findings through publications and presentations in leading journals and ML conferences.

We're hiring one person for this role, and they can be based in either London or Boston. Apply to the posting for the city you'd work from. You only need to apply once.

• Now until October 19th: Accepting applications
• October 26th: Planned start for interviews

• Proven track record building and shipping ML systems with a research component: PhD, publications, or equivalent evidence of independent work, with fluency in PyTorch, TensorFlow or another deep learning framework.
• Experience with software engineering best practices, preferably in Python. You write code others can read and run.
• Foundation model work: pretraining, transfer learning or fine-tuning.
• A collaborative operator. You can hold a real conversation with a biologist about experimental design, take a bioinformatician's pipeline seriously as a dependency, and explain a modeling decision to people who don't build models. You've worked in a small cross-functional team.
• Comfort in a fast-moving, resource-constrained environment where you define your own problem.
• Nice to have: omics, sequencing or molecular data; representation learning over sequences, graphs or chemical structures; small-n, high-dimensional or noisy real-world data, from biology, health or another domain where the data doesn't behave; generative models for molecules or biological sequences; cloud training infrastructure and GPU orchestration.