About the role
The data team decides what models learn from. At trillions of tokens, sourcing, filtering, deduplication and mixing are research problems in their own right: small decisions about what to keep change what the model can do, what it memorises and what it gets wrong. The team builds the pipelines and runs the experiments that connect data choices to measured capability.
Every model Clexerl trains, from the language models behind Clexerl Go to the multimodal systems shared with Autonomy & Sensing, starts here. Data quality is also a safety and legal matter, so the team works with Parent company legal and security on provenance, licensing and the removal of personal information. Decisions are documented so they can be audited later.
Research engineers own the full path from raw crawl to training shards. Most of the work is large-scale data processing, followed by small-model proxy experiments that estimate which changes deserve a full run. Each dataset ships with a datasheet describing what is in it and what was removed.
What you’ll do
- Build large-scale pipelines for extraction, language identification, quality classification, deduplication and decontamination across web, code and document sources.
- Train quality and domain classifiers, and test them with proxy models to see how filtering changes downstream capability.
- Design data mixtures and curricula, and use small-scale experiments to predict which mixture performs best at larger scale.
- Detect and remove personal information, licensed content and benchmark contamination, keeping an auditable record of each decision.
- Generate and validate synthetic data, measuring where it helps and where it causes model collapse or reduced diversity.
- Write datasheets and analyses that tell researchers exactly what each training set contains.
What you’ll bring
- Strong engineering skills in Python and distributed data processing, using Spark, Ray, Beam or similar frameworks on datasets of many terabytes.
- Experience with text deduplication at scale, including MinHash, suffix arrays or embedding-based methods.
- Working knowledge of language model training, enough to design a proxy experiment and interpret its result.
- Good judgement about data quality that goes beyond heuristics, backed by looking at thousands of examples yourself.
- Awareness of legal and ethical dimensions of training data, including licensing, consent and privacy.
- Ability to work across research and infrastructure and explain trade-offs to both.
Nice to have
- Experience building web crawlers or document parsers for PDF, HTML and scanned material.
- Research on data selection, influence functions or data attribution.
- Background in multilingual data and low-resource language coverage.
- Experience with image, audio or video dataset curation at scale.
Work you might take on
- 01
Rebuild a near-duplicate detection pipeline to run across the full corpus and measure its effect on memorisation.
- 02
Train a code-quality classifier and show how filtering with it changes performance on held-out programming tasks.
- 03
Run a mixture-optimisation study with small proxy models and validate its prediction with one larger run.
- 04
Create a personal-information scrubber with measured recall and precision, plus a review process for borderline cases.
Compensation
$280K – $460K
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.