Senior Applied Scientist- Model Training (Supervised/RL) & Optimization

Every nation has data. Few can protect it. Fewer still can act on it.

Dream is the sovereign AI and national cyber-defense company for governments.

We help nations secure their most critical systems, connect fragmented information at a national scale, and turn their most sensitive data into decisions, all fully sovereign.

This is more than a job. It's a Dream job, where you'll work at a global scale alongside some of the best AI researchers, cyber operators, and government experts in the world.

We're building AI that nations own and control, deployed where almost no one else can operate. ingesting and structuring complex data, and driving practical actions that can literally impact the lives of billions of people around the world. This role helps make that real.

Join a team of exceptional researchers building sovereign AI at the frontier. You will advance how we train, adapt, and optimize our models.

You will develop methods across post-training, reinforcement learning, distillation, and predictive ML, spanning modalities that include audio and speech. That means mastering model architectures and training dynamics, designing sharp data strategies, and turning experiments into reliable capabilities under real-world security, compute, and deployment constraints.

Expect scarce data, noisy labels, and problems with no established playbook. You get the autonomy to choose the approach and the responsibility to make it work.

• Develop training and post-training methods, including reinforcement learning, distillation, and automated optimization, that improve model capability and efficiency.
• Design data mixtures, sampling strategies, and curricula, and investigate how data quality and composition drive performance.
• Run and debug large-scale training experiments, tackling convergence, stability, and compute efficiency.
• Adapt architectures and learning methods to new tasks and modalities, including audio and speech understanding.
• Own the problem end to end – definition, experimentation, and validation – using rigorous benchmarks, ablations, and reproducible results to drive decisions, and partner with engineering through deployment.
• Raise risks early, challenge ideas with respect, take on the hardest problems, and mentor fellow researchers.

• 8+ years of hands-on AI/ML, owning multiple projects from problem definition to real-world impact. Industry and military R&D backgrounds are equally valued.
• Deep expertise in model training, with hands-on experience in one or more of: post-training, reinforcement learning, distillation, or predictive ML.
• Strong research foundations: an MSc or PhD in a relevant field, or equivalent demonstrated research accomplishments.
• Excellent implementation skills: Python, PyTorch or JAX, and experience building and debugging large-scale training and evaluation pipelines.
• Deep command of model architectures and training dynamics, with the ability to diagnose failures and balance quality, stability, and efficiency.
• Experience with data mixing, curriculum design, experiment logging, and ablation studies. Research experience in audio or speech is an advantage.
• Independent ownership, empirical discipline, and clear judgment through complex training cycles and deployment constraints.

If you think this role doesn't fully match your skills but are eager to grow and break glass ceilings, we’d love to hear from you!