About Us
KRAI is a cutting-edge AI infrastructure optimization company, a proven and valuable strategic partner for top accelerator designers, server manufacturers, and cloud providers. We are a Founding Member of the non-profit MLCommons consortium, actively contributing to community research and open-source efforts for AI Systems.
We are looking for exceptional R&D engineers to advance the state-of-the-art in AI accelerator programming (accelerating acceleration). The core challenge? Mapping rapidly evolving AI workloads onto rapidly evolving AI accelerator hardware (next generation accelerators, as well as traditional GPUs), while navigating an infinite space of performance, quality, and cost trade-offs.
Our approach combines rigorous performance engineering with systematic agentic techniques. We aim for results that genuinely surprise even seasoned professionals!
What You'll Do
- Developing and optimizing low-level compute kernels for the latest AI workloads.
- Working across a range of accelerator architectures, including hardware that is years from public release.
- Exploring performance, efficiency, and quality trade-offs.
- Driving full-stack inference optimization: from AI models all the way down to hardware.
- Applying both traditional performance engineering tools and frontier AI techniques to solve complex optimization problems.
- Collaborating with top accelerator designers, server manufacturers and cloud providers to deliver best-in-class performance results.
What We're Looking For
- Advanced degree (MSc or PhD) in Computer Engineering, Computer Science, or Natural Sciences.
- 3+ years of hands-on experience optimizing compute-intensive workloads on accelerator hardware (GPUs, TPUs, NPUs, etc).
- Experience with full-stack AI inference optimization: from models to runtimes to kernels.
- Strong command of performance engineering tools: compilers, debuggers, profilers, simulators, and roofline analysis.
- Workflow automation and reproducibility as first-class concerns.
- Strong communication and collaboration skills.
What We're NOT Looking For
- We do NOT design AI hardware: we optimize software that runs on our customers' hardware.
- We do NOT design AI pipelines: we get down to the nitty-gritty of AI inference.
Why KRAI
- Always at the bleeding edge: working with the SOTA AI models and pre-release accelerator hardware.
- Real-world impact: directly influencing hardware roadmaps and procurement decisions at major technology companies.
- Active contributions to open-source and research: getting high visibility and recognition in the AI Systems community.
- Small well-knit team with deep technical expertise and friendly culture.