Staff / Principal Research Engineer - Data Recipe

Neolix; Sunnyvale (CA); On-site; USD 180,000 - 250,000; Full time

Benefits offered by this job

Equity package

Premium benefits: health, dental, and

Vision Insurance

401(k)

Job summary

Neolix US, the Silicon Valley Physical AI team, seeks a Data Recipe owner to decide what real-world experience matters and design data recipes that improve end-to-end driving models.

You will own the learning-value judgment for the fleet data, including what to collect, how to compose datasets, and how to prove improvements through reproducible experiments.

Qualifications

• Must have hands-on experience at a leading autonomous-driving company.
• Experience in data-centric AI and linking experiments to model outcomes.
• Strong Python and practical depth in PyTorch or JAX; implement sampling and experiments.
• Master's or PhD in CS/ML/Robotics or equivalent research/engineering experience.

Responsibilities

• Lead what-to-collect and what-to-curate decisions across ODDs, routes, interaction types and model failure modes.
• Design versioned pre-training and post-training data recipes with datasets and sampling strategies.
• Develop clustering, filtering, near-duplicate detection, long-tail mining and memorization analysis.
• Run controlled ablations and scaling studies to generalize findings to large training runs.
• Build post-training recipes from expert demonstrations and signals.
• Collaborate with model and evaluation teams on held-out tests and robustness.

Skills

Python

PyTorch

JAX

Reinforcement learning

Multimodal learning

Experimentation

Data-centric AI

Dataset curation

Large-scale data handling

Education

Master's or PhD in Computer Science, Machine Learning, Robotics or related field

Job description

Neolix US is the Silicon Valley Physical AI research team of Neolix, the global leader in L4 autonomous delivery. Neolix operates 30,000+ autonomous logistics vehicles in 300+ cities across 15+ countries, with ~200 million autonomous kilometers driven on public roads.

About the role

As our Data Recipe owner, you will decide which real-world experience matters and design the data recipes that turn it into better models. You will own the learning-value judgment for our end-to-end driving program: what the fleet should collect and why, how datasets are composed and sampled, and how we prove those choices improve a video-in, action-out driving model. This is not a data-labeling operations or pipeline-maintenance role.

Must-have: hands-on experience at a leading autonomous-driving company (e.g., Waymo, Tesla Autopilot, Wayve, Momenta, XPeng, Cruise or Nuro). General tech experience alone does not qualify.

Key responsibilities

• Lead what-to-collect and what-to-curate decisions across ODDs, routes, interaction types and model failure modes, and define the triggers, expert demonstrations and intervention signals the fleet should capture
• Design versioned pre-training and post-training data recipes, including dataset compositions, sampling weights, quality filters and training curricula, in partnership with the end-to-end driving team
• Develop clustering, filtering, near-duplicate detection, long-tail mining and memorization analysis for video, sensor and action-conditioned datasets
• Run controlled ablations and scaling studies to find which data mixtures transfer from small experiments to large training runs under matched compute and data budgets
• Build post-training recipes from expert demonstrations, interventions, task feedback and preference signals
• Partner with model and evaluation teams on held-out test sets, train-test overlap audits and long-tail robustness, connecting offline gains to validated real-world behavior

Qualifications

• Required: experience at a leading autonomous-driving company; general tech experience alone does not qualify
• Hands-on experience in data-centric AI, foundation-model training, reinforcement learning or multimodal learning, with direct ownership of experiments that connect data decisions to model outcomes
• A track record of deciding what data a model needs from its failures and distribution gaps, and proving it with reproducible experiments
• Strong Python and practical depth in PyTorch or JAX; able to implement sampling logic, training experiments and evaluation yourself
• Experience with large-scale dataset curation or training recipes, including data mixtures, filtering, deduplication, clustering or hard-example mining
• Strong experimental judgment: controlled ablations, detecting leakage and confounding, and clear conclusions backed by evidence
• A master's or PhD in Computer Science, Machine Learning, Robotics or a related field, or equivalent research and engineering experience

Preferred

• Pre-training and post-training recipes for large video models, VLMs, VLA models or action policies
• Reinforcement learning, imitation learning, preference-data design, and using expert interventions or task outcomes as learning signals
• Data-flywheel ownership at a deployed autonomy company: fleet triggers, active learning, uncertainty-aware selection and auto-labeling
• Data-mixture optimization, scaling laws, curriculum learning and contamination-aware evaluation
• Publications, open-source work or deeply owned industry systems
• Equity: generous founding-team equity package
• Benefits: premium health, dental and vision insurance, 401(k), and comprehensive Silicon Valley tech perks

Success in this role means better real-world driving behavior per unit of data and compute, not simply a larger dataset or a higher offline score. You will define how real-world experience becomes a training signal for Physical AI.