Machine Learning Engineer

Machine Learning Engineer — Physical AI & Robotics

Location: Bangalore (India) or Remote

Type: Full-time
Experience: 0–2 years, or at least 6 months of industry-quality project experience

This role focusses on Physical AI platform being build by the team

Apply: Google form

About the role
You will join a small, early team working at the frontier of ML for robotics and physical AI. Most of your time will go into understanding new papers, evaluating, and turning the ones that work into product. You will see each piece through, from data and models to the backend, frontend and evals around them.

What you'll do

1. Read recent papers in robot learning, simulation and vision. Implement them, find where they break, and propose improvements.
2. Build data and evaluation pipelines: scene generation, simulated robot rollouts, graders and benchmarks. Most of our work is eval-backed: if we can't measure a change, it doesn't ship.
3. Work across the product as needed: data models and APIs, React frontends, tests.
4. Use agentic coding tools (Claude Code, Cursor, Codex etc.) as a normal part of building.
5. Talk to customers, understand the real problem, and reason from first principles about what to build.

What we're looking for

1. Solid grasp of core ML and data science: model training, losses, generalisation, evaluation, statistics.
2. Ability to read a recent paper, understand it in depth and implement it, then think about what to try next and how it could become a product.
3. Experience running the end-to-end ML workflow, including dataset creation, third-party annotator coordination and review, and continuous evaluation.
4. Experience identifying publicly available models and datasets, such as on Hugging Face, for quick deployments and early testing.
5. Foundations in application development: data models / ORMs, React component models.
6. Hands-on experience building with agentic development tools.
7. Willingness to work on any part of the product: frontend, backend, testing, infra.
8. Ability to write high-quality code, proven by prior work.
9. Strong programming and CS skills, preferably through a BE/BTech undergraduate course, with significant ML projects and coursework.
10. Comfort talking to customers, and the habit of thinking from first principles.
11. High self-drive, self-structure, and the ability to work hard.

Our stack
Python, TensorFlow / PyTorch and Kubeflow on Google Cloud; Isaac Sim / Isaac Lab and MuJoCo for simulation.

Nice to have

1. Experience with simulators or robot learning stacks (Isaac Lab, MuJoCo, ManiSkill, LeRobot).
2. Good clarity on image (preferably video) recognition networks, including standard object detection networks; 3D / scene reconstruction.
3. Experience with end-to-end computer vision: data collection, training, optimization for mobile / on-prem deployment, and observability.
4. Running GPU workloads on cloud VMs.
5. Open-source contributions, publications, or a project others have used.

How to apply
Please apply on this form
Please think through all the fileds in the detailed form. The form also asks you to point us to your favourite project: code, paper, demo or write-up, including

1. What problem it solved and why it mattered to you.
2. Your role. What exactly did you build?
3. The 2–3 key choices you made (model, data, architecture, eval) and the alternatives you rejected.
4. How you measured whether it worked, and what the results were.
5. What you would do differently now. Also include: links to GitHub / portfolio, your CV, the agentic tools you use and one thing they changed about how you work, and a recent paper you found exciting and why.

About the product
The role will focus on building in physical AI data science platform: a data and evaluation layer for robotics. It turns video, images or text into simulation-ready scenes, runs graded robot rollouts on simulators such as Isaac Lab and MuJoCo, and produces evaluated training data with sim-to-real checks. This is where this role will spend most of its time.