AI Researcher (Perception)
VC-backed Defence Research Lab | London (on-site)
Compensation: £140k-£175k + equity package
Abou
Most perception systems still rely on expensive hardware, specialist supply chains, and technology that has changed surprisingly little in decades.
This is a stealth research lab, backed by leading European investors, building novel perception systems for defence and security – pushing more capability into software and deploying it at the edge.
The team is currently six people and growing. They bring experience across defence, deployed AI systems, security, and hardware acceleration, including real-world deployments at organisations such as Palantir and Helsing.
The work is classical and modern computer vision – not generative AI, not LLMs. They’re building models that run on severely constrained edge hardware, where power, latency, and memory are hard limits. Custom silicon acceleration is part of the stack.
What you'll do
- Own the perception problem end-to-end – architecture, training, evaluation, and iteration
- Build and optimise models for constrained edge hardware
- Apply quantisation, pruning, and compression techniques to hit strict power and latency targets
- Work across a model family spanning compact architectures up to transformer and SSM approaches
- Collaborate with hardware engineers to validate model–accelerator integration
- Design rigorous experiments and translate results into shipped systems
- Help shape the research direction as the lab scales
What you'll need
- Strong background in computer vision and/or edge ML – classical and modern methods
- Hands-on experience optimising models for constrained hardware: quantisation, pruning, distillation
- Solid Python and deep learning framework experience (PyTorch or similar)
- Track record of building and shipping AI systems, not just publishing
- Genuine interest in defence and security as a mission area
- Applied mindset – you work at the edge of what's possible and deliver results
Bonus
- Experience with transformer or SSM (Mamba) architectures
- Familiarity with hardware-accelerated inference or custom silicon pipelines
- Background in signal processing or low-level sensor integration
Shortlisted candidates will be contacted within 48 hours.