About the role
The multimodal team trains models that read text, images, audio and video as one picture of the world rather than as separate inputs stitched together. The research questions are about representation: how modalities should be tokenised, how they should be aligned, and how a model should reason across them without falling back on whichever one is easiest.
Multimodal understanding matters to every Clexerl division. Clexerl Go agents need to read screenshots, documents and recorded meetings, and the same perception work informs Autonomy & Sensing, where cameras sit alongside lidar and radar. The team shares benchmarks and datasets across divisions so progress in one place is not rediscovered in another.
Scientists here work from the data upwards. You will spend real time looking at examples where the model is confidently wrong, such as a chart read backwards or a video event placed in the wrong order, and designing training changes that fix the class of error.
What you’ll do
- Research architectures and tokenisation schemes for images, audio and video, including learned codecs, patch-based encoders and interleaved sequence formats.
- Develop pretraining and post-training recipes that give a single model strong cross-modal grounding without degrading text ability.
- Study temporal reasoning in video: event ordering, counting, long-range dependencies and efficient frame sampling.
- Build evaluations for document, chart and screen understanding that resist shortcut solutions.
- Analyse hallucination in perception, such as describing objects that are absent, and test mitigations.
- Work with the data team on caption quality, synthetic data and alignment between modalities.
What you’ll bring
- A research record in vision-language models, multimodal learning, self-supervised representation learning or a related field.
- Deep familiarity with contrastive and generative objectives, vision transformers and cross-attention fusion.
- Experience training models at meaningful scale, or enough ablation work to show you understand what transfers.
- Sound experimental discipline when benchmarks are noisy, leaky or poorly specified.
- Comfort with PyTorch or JAX and with efficient handling of large image and video datasets.
- Ability to explain a failure case from first principles, not only report its frequency.
Nice to have
- Experience with diffusion or flow-based generative models for images or video.
- Work on OCR-free document understanding or GUI grounding.
- Background in 3D vision, geometry or sensor data that links to perception research elsewhere at Clexerl.
- Research on efficient long-video modelling with token compression.
Work you might take on
- 01
Compare three video tokenisation schemes on long-clip question answering and publish the compute and accuracy trade-offs.
- 02
Build a screen-understanding benchmark from realistic workspace interfaces and measure where current models misread layout.
- 03
Investigate why a model ignores the image when text priors are strong, and train against that shortcut.
- 04
Develop an interleaved training mixture that adds audio understanding without regressing text and vision scores.
Compensation
$340K – $560K
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.