Job description

WHO WE ARE Odyssey odyssey.ml is an AI lab pioneering general world models: causal, multimodal systems that learn to predict and interact with the world over long horizons. This foundational technology promises to revolutionize robotics, science, healthcare, education, gaming, defense, and beyond. Odyssey’s founders previously pioneered the most complex application of physical AI: self-driving cars. They’ve now brought together a world-class research team from DeepMind, Tesla, Waymo, Meta, Apple, and Wayve, who have made significant contributions to language models (DeepMind Gemini), video models (DeepMind Veo), world models (Wayve GAIA), and autonomous systems (Tesla FSD). Odyssey has raised significant venture capital from GV, Amazon, AMD, EQT, NVIDIA, Natural Capital, In-Q-Tel, Elad Gil, Jeff Dean, Guillermo Rauch, Garry Tan, Kyle Vogt, and researchers from OpenAI, DeepMind, MSL, Recursive, and Thinking Machines. WHAT WE'RE LOOKING FOR Data is fast becoming one of the biggest bottlenecks in building world models. Our models are only as good as the data behind them, and getting that data right is one of the hardest and most important problems we have. We're looking for a data engineer who wants to be the person figuring it out: building and curating the large-scale multimodal datasets our models train on, across video, robotics, and audio. The work spans research and infrastructure. Some days you'll be tuning the platform that processes millions of hours of video and audio. Other days you'll be in the data itself: pulling out signals, improving captions, and working with researchers on what goes into the training mix. We don't treat those as separate jobs, so we want someone comfortable doing both. We care more about how you think about data than about your years of experience or publication record. You might have started in data engineering and moved toward research, or started in research or data science and moved toward engineering. Either way, you like working on data and you've done hands-on work with video and/or audio. WHAT YOU'LL DO - Build and run data pipelines for large-scale multimodal datasets, from raw video, robotics, and audio through to curated, enriched training data. - Pull useful signals out of raw data: detecting speakers, isolating background audio, tracking points and features in video, and other signals that affect what the models learn. - Work on the training mix. Dedup, rebalance clusters, and dig into what's actually in the data so we can find gaps and keep it balanced. - Work directly with researchers, turning model requirements into a data strategy and their questions into pipeline work. - Improve the platform: storage, database layers, throughput, and deployment, so the whole system runs faster and more reliably. WHO YOU ARE - You're good at figuring things out. Give you an ambiguous, poorly-defined data problem and you'll work out what actually matters, then go build it. This is the trait we care about most. - You like the data work itself. Digging through messy multimodal data to find signal is the interesting part for you, not a means to an end. - You can switch between platform and infrastructure work and research-facing feature work, and you see how the two feed each other. - You've worked with real video and/or audio data and know what it takes to process it at scale. - You're comfortable working alongside researchers, translating what they need into concrete data work. - Roughly 1 to 5 years of relevant experience. We're open on seniority and hire for the person, not the title. BONUS POINTS - Computer vision experience, including classic CV like optical flow and feature or point tracking, not only deep models. - Experience with large-scale video or multimodal pre-training, data mixes, and curating data at that scale. - A background that mixes statistics or data science with engineering, plus hands-on video and image analysis. - Exposure to robotics data, human-movement data, or other emerging modalities like wearables or audio-heavy domains. - Audio experience specifically, which is hard to find and we value highly. Our staff is expected to be in office at least three times a week. A lot of our progress comes from working through problems together, and we want you in the room for that.

Requirements

Must-haves

Build and run data pipelines for large-scale multimodal datasets, from raw video, robotics, and audio through to curated, enriched training data.
Pull useful signals out of raw data: detecting speakers, isolating background audio, tracking points and features in video, and other signals that affect what the models learn.
Work on the training mix. Dedup, rebalance clusters, and dig into what's actually in the data so we can find gaps and keep it balanced.
Work directly with researchers, turning model requirements into a data strategy and their questions into pipeline work.
Improve the platform: storage, database layers, throughput, and deployment, so the whole system runs faster and more reliably.
You're good at figuring things out. Give you an ambiguous, poorly-defined data problem and you'll work out what actually matters, then go build it. This is the trait we care about most.
You like the data work itself. Digging through messy multimodal data to find signal is the interesting part for you, not a means to an end.
You can switch between platform and infrastructure work and research-facing feature work, and you see how the two feed each other.
You've worked with real video and/or audio data and know what it takes to process it at scale.
You're comfortable working alongside researchers, translating what they need into concrete data work.
Roughly 1 to 5 years of relevant experience. We're open on seniority and hire for the person, not the title.

Nice-to-haves

Computer vision experience, including classic CV like optical flow and feature or point tracking, not only deep models.
Experience with large-scale video or multimodal pre-training, data mixes, and curating data at that scale.
A background that mixes statistics or data science with engineering, plus hands-on video and image analysis.
Exposure to robotics data, human-movement data, or other emerging modalities like wearables or audio-heavy domains.
Audio experience specifically, which is hard to find and we value highly.
Our staff is expected to be in office at least three times a week. A lot of our progress comes from working through problems together, and we want you in the room for that.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

mlaigo

Platform: ashby
Effort: quick

Benefits

Benefits detected in the listing.

Vision