Job Description
Contract Type: 6 month contract outsourced via agency on an hourly rate
Location: Egham
Hybrid: 3 days onsite (minimum) and 2 days working from home
Rate: Very much dependant on level of experience.
Key responsibilities include:
* Performance Optimization: Profile and debug performance bottlenecks at the OS, runtime, and model levels.
* Model Deployment: Work across the stack—from model conversion, quantization, and optimization to runtime integration of AI models on-device.
* Toolchain Evaluation: Compare deployment toolchains and runtimes for latency, memory, and accuracy trade-offs.
* Open-Source Contribution: Enhance open-source libraries by adding new features and improving capabilities.
* Experimentation & Analysis: Conduct rigorous experiments and statistical analysis to evaluate algorithms and systems.
* Prototyping: Lead the development of software prototypes and experimental systems with high code quality.
* Collaboration: Work closely with a multidisciplinary team of researchers and engineers to integrate research findings into products.
* We not require a PhD holder this time which is unusual for the AI Team.
We're looking for someone with:
o Technical Expertise: Strong OS fundamentals (memory management, multithreading, user/kernel mode interaction) and expertise in ARM CPU architectures.
o Programming Skills: Expert proficiency in Python and Rust, with desirable knowledge in C and C++.
o AI Knowledge: Solid understanding of machine learning and deep learning fundamentals, including architectures and evaluation metrics.
o Problem-Solving: Strong analytical skills and the ability to design and conduct rigorous experiments.
o Team Player: Excellent communication and collaboration skills, with a results-oriented attitude
Desirable Skills:
o Experience with ARM 64-bit architecture and CPU hardware architectures.
o Knowledge of trusted execution environments (confidential computing).
o Hands-on experience with deep learning model optimization (quantization, pruning, distillation).
o Familiarity with lightweight inference runtimes (ExecuTorch, llama.cpp, Candle).