Research Engineer (Physical AI - Models and/or Robotics)

Gtm Week; Toulouse; Hybrid; EUR 60,000 - 90,000; Full time

Benefits offered by this job

Equity: 0.1% - 0.15% stock options

Compute budget and conference travel

Location: Toulouse, France (on-site)

Job summary

TechTree is opening a research lab in Toulouse, France, to advance data-driven robotics and AI. You’ll join a team building the lab’s infrastructure, enabling researchers to run large-scale experiments with multi-GPU training, data pipelines, and real-world deployment capabilities in a fast-paced, globally distributed environment.

You will work on training stacks, data generation, reproducibility, and tools that accelerate research and bring results to production.

Qualifications

• Engineering degree or Master’s from a top programme; PhD a plus, not required.
• 2+ years as a research/ML engineer implementing papers in a team.
• Fluent in English (written and spoken).
• Experience training large models and handling distributed training, checkpoints and evaluation.
• Able to turn papers into runnable code and spot implausible results.
• Expertise in at least one of: distributed training/inference optimization; simulation; robotics software; large-scale video/3D data.

Responsibilities

• Build and run the Lab’s training stack: multi-GPU/multi-node training, experiment tracking, evaluation harnesses and compute budget.
• Turn raw multi-camera data into clean, versioned datasets and generate simulation data as needed.
• Reproduce papers quickly to validate ideas on our data within days.
• Design end-to-end experiments with researchers and identify when results are misleading.
• Deploy models to teleoperation lab and capture hardware, enabling practical tests.

Skills

Python

PyTorch

English fluency

Experimentation

Education

Engineering degree or Master's

Tools

Linux

Slurm

Kubernetes

Docker

CUDA

JAX

Job description

We are a fast-growing, Y Combinator-backed data company headquartered in San Francisco, collecting human and robot data to train robots. We already work with some of the most recognised robotics and AI labs in the United States, and we plan to raise our Series A before the end of the year.

Today we specialise in collecting real-world egocentric data at a scale and speed no lab could reach on its own. Every contributor is identity-verified (KYC), and every contribution goes through several layers of quality control before it reaches a client. Looking ahead, we aim to become a key player in deploying robots at scale and to help make physical AI data accessible.

We are now opening our research lab in Toulouse, France. With a large volume of varied data collected directly at the source by our contributors - access no academic lab has - the Lab will support the future deployment of our robot fleet through research. Its engineers build the infrastructure to turn ideas into results quickly and at scale, and then turn those results into data the teams training tomorrow's models can use.

The role

Researchers decide what to test. You make it run: at scale, reliably and fast. You'll build the Lab's technical backbone and connect research, our data pipeline and the hardware of our lab in Toulouse.

What you'll do

Training stack: build and run the Lab's training stack - multi-GPU and multi-node training, experiment tracking, evaluation harnesses and the compute budget.

Research datasets: turn raw multi-camera recordings into clean, versioned datasets, and generate simulation data when needed.

Fast reproduction: reproduce papers quickly, so the team knows within days whether an idea holds on our data.

End-to-end experiments: design runs with the researchers, catch the bugs that quietly distort a result, and say what the numbers show - and what they don't.

Real-world deployment: bring models into our teleoperation lab, robotic cells and capture hardware.

Open science: publish reproducible code, datasets and models, and co-author the papers you make possible.

Tools for the company: turn the Lab's results into tools for the rest of the business, such as a hand-pose or depth model in our data pipeline.

What we're looking for

An engineering degree or master's from a top programme. For French candidates, a grande école; the MVA master's is a strong signal. A PhD is a plus, not a requirement.

2+ years as a research engineer or ML engineer in a team where implementing papers is part of the job. Exceptional junior profiles are welcome.

Excellent Python and PyTorch, comfortable on GPU clusters (Linux, Slurm or Kubernetes, Docker). JAX is a plus.

You have trained large models and know what breaks: data loading, distributed training, mixed precision, checkpoints, evaluation.

You can turn a paper into code that works, and you spot a result that is too good to be true.

Expertise in at least one of: distributed training and inference optimisation; simulation (MuJoCo, Isaac or other physics engines); robotics software (ROS, sensors, calibration); large-scale video and 3D data.

You like making other people's work faster, and researchers like working with you.

Fluent English, written and spoken. Interviews are held in English.

Nice to have

Co-author of papers at major ML or robotics conferences.

Open-source contributions, or a Hugging Face profile with models or datasets.

Robotic hardware or embedded deployment: quantisation, ONNX, TensorRT.

Egocentric video, depth, hand pose, IMU.

Stack

PyTorch, JAX, CUDA, multi-GPU training on H100s, Hugging Face.

Multi-stream video with depth and IMU, hardware-synchronised and calibrated, delivered in MCAP and LeRobot formats.

Postgres, S3 and Kubernetes on the data side.

Simulators and robotics software as the research requires.

Data no academic lab has, and a team that collects more when the research calls for it.

A founding seat: you choose the tools and set the Lab's engineering standards.

Publication and open source by default, with clear rules on what stays private, agreed before you join.

Your work goes straight to the biggest AI and robotics labs, as data, models and benchmarks they use.

A San Francisco company with a research Lab in Toulouse and a team across San Francisco, São Paulo and Toulouse.

What we offer

Equity: 0.1% - 0.15% in stock options

Your own compute budget and conference travel

Location: Toulouse, France (on-site)

Hiring process

Profile review - papers and citations, experience and education

Culture-fit call with our Head of Recruitment (30 min)

Take-home exercise designed by our Head of Research, based on our data

Call with our Head of Research (1 h)

Call with our CEO (30 min)