London or Bristol, 3 days in the office, 2 days WFH

At Fractile, we’re building what we believe will be the world’s fastest AI inference chip from the ground up. We're balanced across hardware and software engineering, and HW/SW co-design is real here. We move fast, and we help each other move fast. We care about each other, the software we ship and the people who rely on it.

This role sits at the boundary between host and silicon. The kernel driver is key to keeping pace with our ultra-fast devices on cutting-edge server platforms. It’s a high-leverage layer where each win shows up as real throughput and latency gains.

You’ll be there for the pre-silicon simulations, first bring-up, first end-to-end runs, and the moments where performance jumps because of something you shipped.

What you’ll do

• Design, develop, and maintain the Linux kernel driver for Fractile’s AI accelerator
• Build the performance-critical kernel-space paths that keep the host stack moving: PCIe device management, DMA, memory handling, interrupt handling, and robust recovery paths
• Define and evolve the userspace interface the runtime depends on, with a focus on stability, debuggability, and performance
• Work in a tight co-design loop with hardware, firmware, runtime and ML engineers: validate assumptions, iterate on behaviour, and close gaps early
• Improve tooling/workflows so we move fast without breaking things

What we’re looking for

• You’ve shipped and maintained Linux kernel device drivers in production. Experience with PCIe-attached devices and custom hardware interfaces is ideal.
• Strong systems and performance instincts. You can reason about computer architecture, concurrency, and where the time goes
• Excellent C and/or Rust, and a pragmatic approach to building high-quality, maintainable & secure kernel code
• You enjoy hard kernel and system-level problems, and you take them end-to-end until the measurements move
• Computer Science, Electronic Engineering, Maths, Physics, or related degree and 3+ years of industry experience

Nice to have

• GPUs / ML accelerators / high‑throughput I/O (NICs, RDMA, storage)
• Hardware bring‑up or pre‑silicon simulation experience
• Performance work: profiling, tracing, benchmarking, regression tracking

If you want to build the software where every driver win unlocks huge system performance, come build it together.