World Model Algorithm Engineer – Vision & Multimodal Foundation Models

Location: Shenzhen, China

Employment Type: Full-time

Department: Applied Research

About XGEN Labs

XGEN Labs is a Shenzhen-based frontier AI lab focused on World Models.

We are exploring a new paradigm beyond content generation: building AI systems that can understand, simulate, and generate dynamic worlds with persistent states, multimodal perception, and interactive experiences.

Our research spans World Models, multimodal foundation models, video generation, agents, and AI systems. We are looking for exceptional builders and researchers who want to turn ambitious ideas into real systems.

About the Role

We are looking for a World Model Algorithm Engineer to work on vision and multimodal foundation model post-training.

You will develop and optimize post-training strategies for multimodal and video models, using SFT, OPD, RL, and related methods to improve generation quality, instruction following, visual understanding, and temporal consistency.

You will work closely with research and engineering teams to build a continuous data → training → evaluation → iteration loop for next-generation World Models.

What You'll Do

• Develop and optimize post-training pipelines for vision and multimodal foundation models.
• Apply SFT, OPD, RL and related post-training methods to improve model capabilities.
• Design training strategies for video generation, visual understanding, instruction following, and temporal consistency.
• Build, curate, filter, and analyze training data, including hard-example mining.
• Develop evaluation datasets, metrics, and feedback loops to continuously improve model performance.
• Diagnose training degradation, overfitting, instability, and other model optimization issues.
• Continuously improve training recipes and experimental methodology.
• Collaborate closely with researchers and engineers to translate research ideas into scalable World Model systems.

What We're Looking For

• Master's degree or above in Computer Science, Artificial Intelligence, Machine Learning, or a related field.
• Hands-on experience with post-training of vision, multimodal, or video foundation models.
• Strong understanding of SFT, OPD, RL or related post-training paradigms.
• Experience with training data curation, hard-example mining, and model evaluation.
• Strong ability to diagnose and resolve training quality, stability, and optimization issues.
• Strong PyTorch skills and practical deep learning experience.

Preferred Qualifications

• Experience with World Models or video generation.
• Experience with DiT, diffusion models, or autoregressive video models.
• Experience with multimodal generation, video understanding, or temporal consistency optimization.
• Experience building end-to-end training and evaluation pipelines.
• Experience working in a frontier AI lab or foundation model team.
• Strong research mindset and ability to turn ideas into working experiments.

Why XGEN Labs

• Build next-generation World Models at the frontier of AI.
• Work closely with researchers and engineers across multimodal AI, agents, and AI systems.
• Access to substantial compute and modern AI development tools.
• Flexible working environment with no mandatory time-clock punching.
• Five social insurance + housing fund fully contributed; housing fund contribution rate of 12%.
• Two-day weekends, 12 days of annual paid leave, and 2 days of paid sick leave per month.
• Complimentary lunch and dinner, coffee, gym, and recreational facilities.
• Clear career development and growth opportunities.

Location: Shenzhen, Nanshan

We are looking for people who believe the best model is still the next one to build.

Join XGEN Labs and help us build the next world.