Iterative Labs
02Careers

Build the lab that trains the model.

Iterative Labs builds self-driving laboratories and the reinforcement-learning environments around them, so models learn from real experiments instead of text about them. We are a small team backed by Y Combinator, with the first bench running in biology.

Software Engineer

Engineering · Full time or internship

The role

You will work across the whole stack: the reinforcement-learning environments, the simulators and lab hardware behind them, the data that comes back from real runs, and the tooling for training and evaluating models on it. There is no narrow lane. You will own problems from the bench to the model.

What you will do

  • Build environments end to end, from task specification to graded rollout.
  • Put simulators and lab hardware behind one interface, with every run logged and reproducible.
  • Own the data: schemas, storage, versioning, and the tooling the science team uses every day.
  • Build the training and evaluation loop: parallel rollouts, batched evaluation, recovery from failed runs.

What we look for

  • Strong fundamentals and range. You have shipped software that other people depended on, and you pick up new layers of the stack quickly.
  • Comfort with distributed jobs, containers, and data that arrives messy.
  • Experience with RL environments, evaluation harnesses, model training, or lab automation is a plus. Interns welcome.
  • Curiosity about the science. You will work beside the people running the experiments.

Describe one thing you built or led where the impact can be measured. Three or four sentences: the problem, what you did, and the result with numbers. Start with a verb and include technical detail. This answer carries more weight than anything else in your application.

Domain Scientist

Science · Full time or part time

The role

You will decide what the model should learn. That means choosing scientific tasks worth solving, writing the protocols an automated bench can run, judging whether a result is real, and turning all of it into environments a model can be trained on.

What you will do

  • Define tasks with measurable outcomes, starting in biology with protocols such as cell-free protein expression.
  • Design and validate protocols for automated execution, and decide what counts as a successful run.
  • Review what models and agents produce, and identify where they fail and why.
  • Work with the engineering team to encode your judgment into graders and environment design.

What we look for

  • A PhD or equivalent hands-on experience in a wet-lab discipline: biology, chemistry, or materials.
  • You have run experiments yourself and know how they fail.
  • Working experience using language models in your own research, and opinions about where they fall short.
  • Clear writing. Much of the job is turning expertise into precise specifications.

Where do today’s language models and agents fall short in your scientific workflow? Give one concrete failure you have seen, name the models or tools you were using, and say what you would change in the model, the data, or the environment to fix it. Three or four sentences.