3+ years in ML engineering or research, with flexibility for exceptional candidates from frontier labs or AI-native companies.
Hands-on experience enhancing LLM or agent systems through evaluations and post-training methods.
Strong engineering skills, capable of writing pipelines, training scripts, and grading systems independently.
Ability to design experiments that yield reliable results and discern real differences in data.
Comfortable selecting experiments in uncertain conditions, demonstrating adaptability and creativity.
Daily use of modern AI coding agents and a commitment to continuous learning and improvement.
Responsibilities
Identify failure points in agent performance using real hardware data and custom tasks.
Transform identified failures into evaluations, environments, and training datasets based on actual test outcomes.
Develop grading and verification systems to assess agent performance against empirical data.
Implement fixes based on findings, including prompt adjustments, tool modifications, and model fine-tuning.
Prototype innovative research initiatives linking hardware contexts, engineering tools, and simulations to physical tests.
Establish the research infrastructure necessary for efficient experimentation, including evaluation harnesses and data pipelines.
Differentiate genuine improvements from noise, integrate successful changes into customer-facing agents, and document findings.
Benefits
100% coverage of medical, dental, and vision insurance.
Unlimited PTO and sick leave.
Free lunch, snacks, and coffee provided in the office.
Professional Development Stipend to support career growth.
Access to an in-office hardware lab with a $250 project stipend.
Annual company retreat for team building and relaxation.
12 weeks of parental leave for new parents.
Fertility and family planning support up to $15,000 lifetime.
Wellness stipend of $100 per month to promote health.
$500 per year hardware sensor stipend for personal projects.
401(k) plan to support retirement savings.
Employee-friendly equity terms to share in company success.
In-office work required 5 days per week.
Full Job Description
What you'll do
Find where our agents fail on real hardware work, from customer traces and from tasks you design with hardware experts.
Turn those failures into evals, environments, and training data grounded in real test results.
Build graders and verifiers that check agent work against measured data, and find where they break.
Fix what you find: change a prompt, tool, or context, fine-tune, run RL, or distill an open-weight model.
Prototype across our research bets: hardware context for agents, computer use in engineering tools, linking simulation to physical tests, coordinating collections of agents, and intelligence at the edge.
Build the research infrastructure the loop runs on: eval harnesses, dataset and trace pipelines, and training and experiment tooling that make the next experiment faster to run and easy to reproduce.
Separate real gains from noise, ship what wins into the agents customers use, and write up what you learn.
What you'll bring
3+ years in ML engineering or research. Less is fine if your recent work was doing exactly this at a frontier lab or an AI-native company.
Hands-on experience improving an LLM or agent system through evals, data, post-training, or harness changes, and proving it worked.
Strong engineering: you write the pipeline, the training script, and the grader yourself.
Experimental judgment: you design experiments that don't fool you, and you know when a difference is real.
Comfort picking the next experiment when nobody can tell you what will work.
You build with modern AI coding agents (Claude Code, Cursor, Codex) every day, and stay curious and open to better ways of working. The tools keep changing, and so do we.
Nice to have
You've done post-training (SFT, RL from feedback or verifiable rewards, distillation) on real tasks.
You've built RL environments, synthetic-data pipelines, or benchmarks that other people use.
You've worked on computer-use agents, multi-agent systems, simulation, or edge deployment.
You've worked with time-series, sensor, or scientific data, or with models built over it.
You've been an engineer in hardware, test, controls, or robotics, or worked closely with those teams.
Benefits/Perks
100% coverage of medical, dental, and vision insurance
Unlimited PTO and sick leave
Free lunch, snacks, and coffee
Professional Development Stipend
In-office hardware lab with a $250 project stipend
Annual company retreat
12 weeks parental leave
Fertility and family planning ($15,000 lifetime)
Wellness stipend ($100/month)
$500/year hardware sensor stipend
401(k)
Employee-friendly equity terms
In-office 5 days per week
Compensation
The base pay range for this role is $230,000 - $270,000 per year.
