Machine Learning Engineer Global Public Sector

## About the role Scale's mission is to develop reliable AI systems for the world's most important decisions. Our core work consists of: - Creating custom AI applications that will impact millions of citizens - Generating high-quality training data for national LLMs - Upskilling and advisory services to spread the impact of AI Scale is hiring ML Research Engineers to bridge the gap between emerging AI capabilities and mission-critical, real-world impact. In our Global Public Sector (GPS) division, we don't just implement tools; we conduct applied research to solve the unique challenges of sovereign AI. Your role is to move beyond off-the-shelf implementations. You will lead the research into Agent Design, Reliability, and AI Safety, developing novel system architectures that power high-stakes government applications. You will be the bridge between a research paper and a production-ready system that functions at scale. ## The Mission - **Applied Agent Research:** Leading the design of reliable, multi-step agentic systems and long-horizon reasoning frameworks that can solve complex problems for national security and public policy. - **Systemic Evaluation & Red-Teaming:** Developing rigorous benchmarks and evaluation protocols to ensure AI systems are safe, unbiased, and performant in high-stakes, non-commercial environments. - **Model Optimisation & Selection:** Conducting deep-dive research into model performance (both open-weight and closed) to identify the best tools for niche domains, optimising them through context engineering, RAG, and other inference-time techniques. ## What You Will Do - **Architect Agentic Systems:** Design and build agent architectures, the harnesses, tool-use protocols, and logic flows that allow LLMs to function as reliable, autonomous agents in complex workflows. - **Drive Reliability & Safety:** Research and implement robust evaluation frameworks. This includes red-teaming for sovereign AI requirements and developing strategies to mitigate hallucinations in regulated data environments. - **Synthesise Deep Research:** Build agents capable of autonomous information synthesis and long-horizon reasoning, enabling users to analyse massive datasets and extract actionable insights. - **Optimize for Niche Domains:** Evaluate and adapt models for specialised use cases, such as LLM reasoning for low-resource languages, complex OCR tasks, or working in GPU-constrained environments. - **Build Evaluation Frontiers:** Create new, automated benchmarks that define what success looks like for AI in the public sector, ensuring our systems meet the highest standards of accuracy and sovereignty. - **Consult as a Technical Authority:** Act as a subject matter expert for public sector leaders, advising on the practical limits, safety requirements, and performance trade-offs of emerging AI technologies. ## Ideally, You Have - **Engineering Rigour:** Exceptional proficiency in Python and experience building agentic harnesses or AI infrastructure. You write production-ready code that is modular, scalable, and reliable. - **Applied Research Mindset:** A track record of taking theoretical AI concepts and turning them into functional prototypes or products. You know how to read a paper and determine if its methods are actually viable for a production system. - **Evaluation Expertise:** Experience in LLM benchmarking, red-teaming, or building evaluations that go beyond standard academic datasets. - **Advanced Degree:** A Master's or PhD in Computer Science, Mathematics, or a related field (with a focus on ML) is preferred, but we value demonstrated impact and engineering excellence. ## Nice to Haves - **Agentic Systems Expert:** Deep experience in building multi-agent systems, including chain-of-thought optimisation and tool-calling reliability. - **Sovereign AI Experience:** Experience working with highly regulated data environments, on-premise deployments, or sensitive government use cases. - **Inference Optimisation:** Knowledge of how to optimise model performance for environments with limited GPU capacity or specific latency requirements. - **Zero-to-One Mindset:** You are comfortable navigating ambiguity and enjoy defining research directions from scratch to solve a specific product or mission need.