Research Engineer

About the role

About Roam

Roam is an Applied AI lab building World Models for highly retentive 3D environments. We're lifting models from the 2D plane of pixels into the full dimensionality of interactive worlds. Our work spans everything required to perceive, reason about, and create in 3D, with applications across simulation, design, and play.

Our first focus is gaming: 3D environment generation and understanding for interactive entertainment, technology that enables millions of people to explore, create, and share highly engaging worlds together.

Similar to how TikTok turned everyone into a creator through easy-to-use tools, we will let everyone turn their ideas into games without needing any technical skills.

Since founding earlier this year, we have:

- Locked in 1,000+ gaming & livestream creators (100 K–10 M followers each) who will launch Roam games to an aggregate reach of 250 M+ fans.

- Assembled a 12 person applied AI team spanning Ex-Moonshot (post-training & agents, diffusion model training), second-time technical founders, engineers that made 100+ games for Voodoo, Supersonic, Sidemen and built mobile apps from 0–2M DAUs.

- Raised an unannounced round from investors behind Google Deepmind, Bytedance, Anduril, Together AI and Polymarket, angels from OpenAI, Anthropic, xAI, Deepmind, Cognition, & large content creators.

- Acquired a consumer product studio with 11.5M+ downloads.

About the Role:

You’ll work at the intersection of applied AI and large-scale systems, focusing on world generation pipelines, data-efficient training methods, and inference-time optimization for real-time generation. Your work will enable everyone make, play and monetize highly engaging games & 3D environments. Our current research spans:

- Distributed multi-agent orchestration systems for coordinated world synthesis, spanning terrain generation, asset population, and behavior trees

- Reinforcement learning pipelines for adaptive, open-ended game generation with hierarchical action spaces and dynamic difficulty adjustment

- Agentic codegen & diffusion-based tooling for creating studio quality simulated environments

We’re looking for engineers who have:

- Strong software fundamentals and comfort switching between production systems and research

- Experience building or training large multimodal or agentic systems

- Familiarity with distributed training, data curation, or large-scale vision/3D pipelines at scale

- Bonus: Have worked on generative vision, world models, or RL systems.

Strong candidates may also have:

- Experience training or fine-tuning diffusion or other generative vision models.

- Prior work in robotics, multi-agent systems, or simulation-based RL.

- Familiarity with game engines or 3D simulation frameworks as experimental testbeds.

Compensation: