About the role
The Data Engine decides which of the many terabytes the fleet records are worth keeping, labelling and training on. You will build the loop that finds rare and informative moments, such as a cyclist behind a truck or a faded lane at dusk, routes them to labelling, retrains models and checks whether the failure is actually gone.
Perception and Planning teams hit the same wall: the long tail is where models fail, and random sampling rarely finds it. A good data engine raises model quality faster than any architecture change. Retrieval, embeddings and multimodal understanding from Frontier AI feed directly into how we search the fleet, and what we learn about labelling and active learning flows back to that group.
You will own pipelines end to end, from on-vehicle triggers and ingest through curation and training sets to evaluation, with clear lineage for every dataset.
What you’ll do
- Build mining pipelines that surface rare scenarios using model disagreement, embedding search, heuristics and triggers running on the vehicle.
- Design active learning and auto-labelling workflows that combine model pre-labels, human review and quality checks to cut cost per label.
- Maintain dataset versioning and lineage so any model can be traced to the exact data, labels and filters that produced it.
- Develop evaluation sets and slice-based metrics that show whether a mined data batch fixed the targeted failure.
- Improve ingest and storage throughput for fleet logs, balancing cost, retrieval latency and the need for full-fidelity sensor data.
- Partner with perception researchers to turn each failure analysis into a concrete data request and track it to resolution.
What you’ll bring
- Experience building ML data pipelines at scale, preferably for perception, vision or other sensor-heavy models.
- Solid ML fundamentals, including training dynamics, label noise, class imbalance and evaluation under distribution shift.
- Strong Python and fluency with distributed data tools such as Spark, Ray or similar, plus familiarity with object storage and columnar formats.
- Experience with embeddings, similarity search and clustering to find and deduplicate examples in large collections.
- Care about data quality: you audit labels, check for leakage between train and test, and write down what you found.
- Ability to work with labelling teams and researchers and translate between their needs.
Nice to have
- Experience with active learning, uncertainty-based sampling or curriculum design for perception models.
- Familiarity with on-vehicle data triggers and constrained uplink strategies.
- Experience using vision-language models to caption or query driving scenes.
- Background in privacy-preserving handling of camera data, including face and plate redaction.
Work you might take on
- 01
Build an embedding index over fleet camera and lidar frames so researchers can find all scenes resembling a failure in seconds.
- 02
Train a model-disagreement miner and show that retraining on its top results lowers error on a held-out rare-object slice.
- 03
Stand up an auto-labelling pass for lidar tracks, with confidence thresholds that route only uncertain frames to human reviewers.
- 04
Create dataset lineage tooling that reproduces any past training set exactly and diffs it against the current one.
Compensation
$220K – $360K
The US base salary range for this role. Every offer also includes equity in Clexerl and, when it spins out, in the venture you help build. We tell you the reasoning behind your number.
Benefits
Health, fully covered
Medical, dental and vision for you and your dependants, premiums paid in full.
Two kinds of equity
Equity in Clexerl, plus a stake in the venture you help build when it spins out.
22 weeks parental leave
Fully paid, for every parent, however your family arrives.
Rest that sticks
Flexible time off with a four-week minimum we actually check on, plus a company-wide week in December.
Learning and compute
A yearly budget for courses, conferences and books, and compute credits for side projects.
401(k) with a 4% match
Matched from your first paycheque and vested immediately.
How we hire
Apply 5 minutes
Send links to what you have built and a few lines on why this role. No cover letter needed.
Intro call 30 minutes
A conversation with the hiring manager about the work, the team and what you want next.
Work sample 2–4 hours
A paid take-home or a pairing session on a real-shaped problem. You pick the format.
Studio day One day
Meet the team, go deep on your work and ours. Remote or in person.
Offer Within 5 days
A written offer with the full range, the equity and the reasoning behind both.
Opens an email with a short template. Links beat résumés.