R&D Engineer

R&D Engineer — Research Role:
Neural Network Architectures for YOLO

At Ultralytics, we commit to relentless innovation in the AI space and seek team members who resonate with our ambition to produce the world's best YOLO AI models. If you're obsessed with AI, eager to make an impact on the world and thrive in dynamic, high-intensity environments, we invite you to apply for a position on our team.

At Ultralytics, we're on a mission to simplify AI for everyone. As the creators of the world's leading open-source YOLO models and the popular Ultralytics Git Hub repository, we empower millions of developers, researchers, and companies worldwide to build state-of-the-art computer vision applications.

Following our $30M Series A round, we're expanding rapidly across our global hubs in London, Madrid, Shenzhen, and New York. This is an opportunity to join a fast-scaling, high-performance team that's redefining the future of vision AI — where ambition meets impact, and ideas become reality.

We move fast. We build boldly. We execute with purpose. And we do it together.

As R&D Engineer — Research Role:
Neural Network Architectures for YOLO, you'll design and develop the neural network architectures at the core of how YOLO is built. Your work will directly shape the open-source models used by millions of developers across detection, segmentation, pose, and emerging vision tasks.

You'll operate at the intersection of research and production, turning ideas from papers into performant architectures, rigorous experiments, and production-ready model designs. Working closely with our cross-disciplinary R&D team, you'll help define future releases across the Ultralytics ecosystem and contribute to the technical foundations documented in our docs and guides.

This role is ideal for a hands-on researcher who thrives on first-principles thinking, rapid experimentation, and high ownership. You bring deep architectural intuition, strong PyTorch skills, and the drive to build original models that advance the state of the art.

Research & architecture design

• Research, design, and build novel neural network architectures for next-generation YOLO models.
• Develop core components including backbones, necks, heads, attention modules, and efficient building blocks.
• Read, reproduce, and extend cutting-edge papers into working model architectures from scratch.

Experimentation & optimization

• Run rigorous ablation studies and large-scale architecture experiments across standard benchmarks.
• Improve accuracy, latency, and scaling through distillation, pruning, and quantization-aware design.
• Build training recipes and evaluation pipelines for detection, segmentation, pose, and new vision tasks.

Collaboration & ownership

• Partner with R&D and engineering teams to turn research ideas into production-ready open-source models.
• Contribute to foundational model initiatives that support future YOLO and broader vision systems.
• Take ownership of assigned R&D tasks beyond architecture work and solve problems efficiently and correctly.

Core requirements

Skills and experience

• 5+ years of hands-on experience in computer vision and deep learning with architecture design expertise.
• Proven success building models from the ground up, not just fine-tuning pretrained checkpoints.
• Expert-level Python and deep proficiency in PyTorch, including custom layers, modules, and training loops.
• Strong command of convolutions, attention, normalization, optimization, and loss design fundamentals.
• Ability to read a research paper and implement it faithfully and efficiently from scratch.
• Experience with efficiency research such as quantization, pruning, distillation, or neural architecture search.
• Strong portfolio of research contributions through papers, preprints, open-source code, or original model work.
• High ownership, adaptability, and comfort operating in a fast-moving research environment.

Nice to have

• Published research at venues such as CVPR, ICCV, ECCV, NeurIPS, or ICLR.
• Experience with distributed training, mixed precision, CUDA profiling, or model deployment constraints.
• Active open-source presence with architecture-focused repositories and meaningful community adoption.
• Familiarity…