Software Engineer, ML Infrastructure

Job Summary

You'll join Arc's Infrastructure Engineering team and own the data and evaluation of ML infrastructure for Arc's virtual cell models. You'll start with two systems: our Model Performance Dashboard, which benchmarks those models, and the pipeline that turns single-cell data into their training sets.

Responsibilities

• Own how Arc measures the performance of its virtual cell models, starting with the backend of our Model Performance Dashboard, which you'll take over within your first two months.
• Decide with the ML team which evaluation tools they need, then design and build them so researchers can evaluate and compare models themselves, and measure what each dataset version does to model performance.
• Redesign the pipeline that turns single-cell data into training sets for our virtual cell models, then run it with AI agents: agents find, process and curate datasets, and automated quality checks, not manual review, decide whether a dataset version is ready for model training.
• Own the lab-in-the-loop infrastructure that turns model performance results into the case for what Arc's labs should test next, working directly with the science leads who design those experiments.
• Build or buy validation frameworks (tests, evals, QC gates, policies) that let agent-written code and apps ship without line-by-line human review.
• Own the engineering that makes the ML team's training and evaluation run fast, reliably and at scale.

Requirements

• Bachelor’s degree with 8 years of relevant experience, or Master’s degree with 6 years of relevant experience.
• You've owned a production data or ML platform that researchers depend on, end to end (pipelines, storage, versioning), and set its technical direction.
• You design and build systems with AI tools and agents, and you make their output verifiable with evals, QC gates or tests before anyone relies on it.
• You've learned an unfamiliar technical domain well enough to make design calls in it.
• You work directly with stakeholders to find requirements, with no PM in between.
• You make the people around you better through code review, pairing and writing patterns down.
• You've diagnosed and fixed performance problems in Python or PyTorch systems with profilers and measurements: GPU utilization, memory, data loading.
• Work onsite in our Palo Alto office at least 3 days a week.

Preferred Qualifications

• You've worked with single-cell or perturbation data (AnnData/h5ad).
• You've built or run model evaluations: benchmarks, eval harnesses, PyTorch/GPU inference.
• You've run multi-node distributed training (DDP/FSDP, NCCL, SLURM) or raised a training job's throughput.
• You've led an effort with one or two other engineers.

Benefits

• The base salary range for this position is $191,000 to $237,000.
• This position is also eligible to receive an annual discretionary bonus, with the amount dependent on individual and institute performance factors.
• You will be eligible to participate in the Institute’s Compound Cash Grant Program (the “CCGP”).