SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
CuspAI is a frontier AI company focused on using artificial intelligence to discover breakthrough materials for energy, clean water, computing, and carbon capture. The company's founding team comprises world-class researchers in AI, chemistry, and engineering, and they are building an AI-powered 'search engine' for next-generation materials discovery.
You will serve as a Performance Engineer, acting as the vital bridge between frontier AI models and high-performance computing infrastructure. Your primary focus will be optimizing large-scale training and inference workloads to accelerate the materials discovery pipeline and enable researchers to iterate rapidly.
Key responsibilities include:
**Training & Inference Optimization:**
- Analyze model architectures to optimize mathematical operations on hardware, focusing on NVIDIA GPU kernels and collective communication layers (NCCL)
- Implement efficiency techniques such as quantization, model pruning, and distillation to reduce latency for materials discovery APIs
- Streamline data pipelines to ensure high-performance accelerators remain fully utilized during massive training runs
**Profiling, Observability & Cost Management:**
- Use advanced observability tools to trace end-to-end execution of training jobs and identify bottlenecks in Model FLOP Utilization (MFU)
- Build and maintain real-time dashboards tracking system metrics including MFU, memory bandwidth utilization, and cluster uptime
- Monitor AI accelerator utilization (e.g., H100 clusters) to prevent idle time and ensure compute spend translates to scientific progress
- Develop scaling laws and internal tools to predict system behavior as models grow in parameter count
**Interdisciplinary Collaboration:**
- Work closely with AI researchers to adapt model architectures for better hardware compatibility without sacrificing scientific accuracy
- Partner with the infrastructure team to influence future compute cluster design based on performance data
The role is based in Cambridge with flexibility across London, Amsterdam, or Berlin offices, with an expectation of three days per week in-office. Initial onboarding will require several weeks in the Cambridge/London office. Regular travel across Europe and globally may be required for collaboration.
**Requirements:**
- 3+ years of experience in performance engineering, site reliability engineering, or low-level software development focused on high-performance computing (HPC) or AI workloads
- Expert-level knowledge of GPU architecture and experience optimizing CUDA kernels or using libraries like Triton
- Deep understanding of distributed training frameworks (PyTorch Distributed, DeepSpeed, Megatron-LM) and networking protocols
- Proficiency in profiling tools such as Nsight Systems, Nsight Compute, or Weights & Biases
- Builder mindset with ability to work autonomously in early-stage, fast-paced startup environment
- Personal interest in the potential applications of AI in materials science
**Bonus qualifications:**
- Experience with specialized AI hardware (TPUs or alternative accelerators)
- Academic background or personal interest in Materials Science, Chemistry, or Physics
- Knowledge of MLOps practices for model deployment at scale in cloud environments (AWS/GCP/Azure)
- Contributions to open-source performance optimization or machine learning libraries
