Data & Model Quality Engineer - Foundation Model Team
Key Responsibilities
Model Training & Checkpoint Selection
- Run model training across a range of hyperparameter configurations.
- Identify stable, reliable checkpoints to use for evaluation, and document the criteria behind each choice.
- Keep experiments reproducible so results can be compared over time and across datasets.
Evaluation & Statistics
- Evaluate models against the current task setup and produce clear statistics (e.g. success rates, failure modes, variance across runs).
- Define evaluation frameworks, metrics and KPIs for both models and datasets.
- Turn test and field failures into concrete model, data or system improvements.
Data Strategy & Collection Feedback
- Use evaluation results to decide what kind of data needs to be collected next, so the dataset keeps improving where it matters most.
- Own the definition of good versus bad data, and keep it written down, up to date and shared.
- Feed collection priorities back to the Operations team on a regular cadence.
Data Quality Tools, Methods & Operator Enablement
- Build tools, methods and know-how that detect, flag and filter problematic data so the dataset meets our quality standards.
- Write clear collection guidelines, and train operator leads to follow and pass on those instructions.
- Monitor whether quality standards are being met, and adjust the guidelines when they are not.
Evaluation & Data Quality Infrastructure
- Where needed, build infrastructure that makes model evaluation and data quality evaluation as easy and automated as possible.
- Contribute to the pipelines that turn raw multimodal data into training and evaluation datasets.
Collaboration
- Work closely with the Hardware, Operations and System Engineering teams.
- Keep track of advances in robot learning, data curation and evaluation methods, and bring useful ones into the team.
Qualifications
- Model training: Experience training models for robotics control, including hyperparameter tuning.
- Model understanding: Ability to understand a model's architecture and approach, and to use that understanding to tune hyperparameters and improve performance.
- Data quality: Experience going through data, identifying problematic parts and improving its quality, ideally with multimodal data such as images, video, trajectories and sensor logs.
- Environment: Experience working in a fast paced and high output environment.
- Programming: Strong Python; experience with other languages is a plus.
Nice to Have
- Experience with vision-language or vision-language-action models, or other large multimodal models.
- Familiarity with robotics software stacks, ROS1/2 or Jetson-class platforms.
- Experience with distributed training or production ML infrastructure.
- Systems thinking: understanding how model behavior interacts with control, safety and system boundaries.
Soft Skills & Culture Fit
- Ownership of outcomes and a user-centric mindset.
- Comfort in a high-performance, high-accountability environment.
- Ability to turn technical findings into clear, practical instructions for non-engineers such as operators.
- Strong English communication; Japanese is a plus.
Supplementary Materials (optional)
Applicants may provide evidence of:
- Project portfolio or demo links
- Data quality work: tools built, problems found and how quality improved
- Model training and evaluation work, including hyperparameter tuning results
- Operational results beyond PoC or research-only settings