AI Researcher

Key Responsibilities

• This role starts with working directly in DOR's recording and editing features to create the data you'll train on, and shifts toward world model training and the research built on top of it as that data comes together.
• Work directly in DOR's recording and editing features to decide, at the product level, which signals get captured and at what cadence, and build the data your experiments need.
• Define the alignment rules that handle the frame drops and timestamp drift caused by screen, key input, and voice being captured on different clocks, and turn the footage that comes with recorded actions — out of the 200 million videos accumulating each month — into trainable sequences.
• Design and train game world models that take observations and actions and predict the next state.
• Design latent state representations that compress high-bandwidth signals like the screen, learn transitions conditioned on key inputs on top of that state, and experiment with where signals like voice belong — signals that sit closer to player intent than to the state of the environment.
• Roll the learned model forward many steps to see what's coming, and look back at how things would have gone if a different action had been taken in the same situation.
• Define the criteria for measuring how far a prediction can be trusted, and decide what to validate counterfactual predictions against when no ground-truth trajectory exists.
• Experiment with agents that plan actions or learn policies by rolling the future forward inside the model without touching the real game, and explore applications such as in-game elements and NPC generation that only become possible once the dynamics are learned.

Products You'll Build

How We Share Our Research

We encourage putting research out into the world. The company supports conference talks, paper submissions, and open-source releases, and where something collides directly with product competitiveness, we scope that part together.

Compute for experiments is provided by the company, sized to the research in flight.

How We Work

• DOR builds products that customers want, not products the CEO wants.
• We repeat problem definition, hypothesis, execution, measurement, and learning in short 1-2 week iterations.
• AI Researchers do research that doesn't end at a paper -- they follow the models they train all the way to the point where those models touch real users' gameplay.
• We build fast, validate fast, and adjust direction fast.
• We focus on creating real impact based on data and user feedback.
• We value results that connect not only to performance metrics but also to user experience.

Tech Stack

We don't expect experience with every stack. However, we expect you to learn and pick things up quickly.

Qualifications

• Experience designing and training deep learning models yourself, whether in lab research, a course project, or a personal project
• Able to design your own experiments in Python and PyTorch and interpret the results
• Experience reading papers or public code and implementing them yourself to see them run
• Experience turning unstructured data into something a model can train on
• Uses AI coding tools (Cursor, Claude Code, etc.) heavily for experiment pipelines and for writing code
• Able to go beyond research code and fix the product code where the data is created

Preferred Qualifications

• [Top priority] Experience building experiment pipelines or research prototypes yourself with AI tools
• Research experience in world models, latent dynamics, sequence modeling, video generation, reinforcement learning (especially model-based RL), or planning
• Have reimplemented a world model or a sequence · video modeling paper yourself, and written up what reproduced and what didn't
• Published or submitted a paper, or contributed to open-source projects
• Currently in a master's or PhD program in machine learning, or have done research in a lab
• A gamer or someone with a deep understanding of gameplay context

You'd Be a Great Fit If You Are...

• Someone who defines problems that don't have answers yet and digs until they do
• Someone who wants to push their own research agenda on data you can't get at school
• Someone who doesn't separate building the model from building the data it learns from
• Someone who wants their research to live on as papers and public code
• Someone who wants to build game world models on the play data of gamers worldwide

What to Highlight in Your Resume

• Tell us which problems you've solved so far and why you chose them. Course projects and personal research count.
• If you've designed and trained a model yourself, tell us the problem definition, your approach, and the attempts that failed.
• If you've reproduced a paper or public code, tell us what reproduced and what didn't.
• If you've turned unstructured data into something a model can train on, describe it concretely.
• Leave links we can check — papers, open-source contributions, Kaggle, lab internships, personal projects.
• Rather than listing work history and credentials, we'd love to see your problem definition, thought process, actions, learnings, and outcomes.

Hiring Process

The entire hiring process is completed within 7 days. During intensive hiring periods, we'll reach out quickly at each stage.