AI Model Architecture Optimization Engineer R&D
San Mateo, CAFull-timePosted 3mo ago
About the role
Experienced AI Acceleration Engineer to research and develop techniques for accelerating large model architectures (e.g., transformers) and integrate advancements into the PyTorch stack. Requires Python proficiency with desirable CUDA experience.
Responsibilities
- Innovate on AI model components, such as attention blocks, KV-cache strategies, layer streaming, tokenization, layer norms, and more, to improve AI model performance and scalability.
- Optimize and integrate AI acceleration techniques into the PyTorch stack, enabling efficient use across diverse hardware platforms.
- Own & drive features end to end to push the limits of large model architecture, ensuring seamless integration with existing frameworks.
- Benchmark and profile AI models to evaluate performance improvements, ensuring optimal execution on target hardware.
- Write and maintain clean, efficient code in Python, with a focus on integration with PyTorch.
- Leverage CUDA for GPU-based acceleration when necessary, optimizing the attention blocks for maximum performance.
- Work on cross-functional teams to design, implement, and test new features.
Qualifications
- Extensive experience with large AI model architectures, particularly with attention blocks and transformer models.
- Proficiency in Python and hands-on experience with the PyTorch framework.
- Strong understanding of AI acceleration techniques and their application in real-world use cases.
- Familiarity with CUDA for GPU programming is highly desirable.
- Demonstrated ability to optimize complex models for performance across different hardware environments.
- Experience in developing and deploying AI models at scale is a plus.
Skills
Required:PythonPyTorch
Nice to have:CUDA
Benefits
- Medical, Dental, and Vision benefits
- Flexible Paid Time Off, 10 days
- Parental Leave
- 401(k) Plan with company matching
- Snacks and coffee to keep you energized
Last updated Oct 25, 2026. Interested? Apply on the company site.
About OpenInfer
OpenInfer — Inference OS for the Agentic Era
OpenInfer is the Inference OS for the agentic era — run AI agents on the CPUs, GPUs, and NPUs you already own, at the best cost and speed, with no cloud lock-in.
AI infrastructureCPUsGPUsInference OSNPUsbessemer-ai-infrastructure-2026distributed inferenceedge computing
More open roles at OpenInfer
- Senior On-Device AI Inference Performance Engineer3mo agoSan Mateo, CA · Engineering
- Full-Time Engineer: AI Software Engineer3mo agoHybrid (Bay Area, CA)