Science Machine is building agentic AI software for automating bioanalysis workflows. Our platform helps scientists and bioanalysis teams turn complex analytical processes into reliable, repeatable, AI-assisted workflows across data ingestion, analysis, reporting, review, and compliance.
The Research Engineer executes our research projects. The work spans agent systems, how we quantify their performance, and fine-tuning open models. Research direction comes from product feedback; you turn it into shipped results.
You will work close to the science: formulating problems, designing experiments, shipping implementations, curating data, and interpreting results. You will work with biologists, scientific contractors, and the rest of the team, and act as a technical sparring partner on what to build next.
What you’ll do
- Turn research directions into projects: scoping, experiment design, implementation, data, and results.
- Define how we measure system performance: which metrics matter, which proxies are honest, which signals to trust.
- Build the evaluation infrastructure to put those metrics into practice: benchmarks, harnesses, and the tooling around them.
- Develop the research agent stack: memory and in-context learning, test time compute, and models.
- Fine-tune and post-train open models to improve agent performance.
- Work with biologists, scientific contractors, and annotators to build reproducible training and evaluation data pipelines.
- Track what frontier labs are shipping and bring back what’s relevant.
- Experience building ML/AI systems in a research-adjacent context (industry, lab, or PhD).
- Experience building LLM-powered systems: prompts, context engineering, agent architectures.
- Experience working with evaluations and benchmarks, including in tasks where “correct” is ambiguous.
- Familiarity with model training generally, including the data, optimisation, and evaluation work around it.
- Strong engineering fundamentals. Fluent in Python and comfortable across the AI/ML stack.
- Experience running experiments rigorously, in academia or industry. You think about confounds. You can defend your results.
- Experience with training and evaluation data pipelines, including reproducibility and observability.
Nice to have
- Experience in a life-science domain (biology, chemistry, medicine, bioinformatics).
- Post-training experience on LLMs.
- Peer-reviewed publications, or other settings where your ideas were stress-tested.
- Open source contributions to scientific or AI tooling.
- High ownership: you notice what needs doing and carry it through.
- Skeptical of your own results: you assume a good-looking number is wrong until you understand why it isn’t.
- Strong opinions, weakly held: you push back, defend a position, and change your mind when evidence moves.
- Hands-on with the unglamorous parts: data cleaning, contractor coordination, eval annotation, whatever the project needs.
- Reliable under ambiguity: you make progress when problems aren’t yet well-defined.
- Curious about science: you want to learn the bioanalysis domain.
- Clear communicator: you can explain technical decisions and results to engineers, scientists, and founders.
#J-18808-Ljbffr…