The Role
This is a technical ownership role, not a feature-delivery seat. You’ll define how LLM systems are built, evaluated, and operated across a cybersecurity product — from architecture decisions through to production monitoring. You’ll work directly with product and engineering, which means you need to translate between research thinking and shipping constraints without losing either.
About the Product
The platform applies AI to real-world cybersecurity problems — threat detection, privacy, and security-sensitive workflows where model reliability and safety aren’t optional. LLMs are a core part of the product direction, not a bolt-on: the work spans prompt engineering, RAG pipelines, fine-tuning, and inference optimisation in systems where accuracy and latency both matter. This is applied AI in a domain where failures have consequences.
Technology Stack: Python throughout. PyTorch, TensorFlow, Hugging Face, LangChain, and LlamaIndex are all in scope — the role isn’t locked to one framework. The product context is cybersecurity, which adds constraints around privacy, security, and responsible AI that shape how systems are designed. On-device and edge deployment experience is a meaningful advantage given the product direction.
What You’ll Be Doing
- Design and build LLM-based features and workflows end-to-end — architecture, implementation, and iteration
- Lead model selection, prompt engineering, fine-tuning, and evaluation across the product
- Develop rigorous evaluation frameworks for quality, safety, accuracy, and performance — not vibe checks
- Optimise LLM systems for latency, cost, and reliability at production scale
- Integrate LLM capabilities into the broader engineering stack alongside backend and product teams
- Define LLM development standards and mentor engineers building in this space
- Research emerging models, frameworks, and deployment approaches — bring recommendations, not just summaries
- Ensure privacy, security, and responsible AI practices are embedded in how systems are built, not audited after the fact
What We Expect
Must-have
- 6+ years of professional experience in AI/ML, data science, or software engineering with a strong AI focus
- Deep hands-on experience with LLMs, NLP, and generative AI — not just API integration, but understanding what’s happening under the hood
- Practical command of the modern LLM stack: fine-tuning, prompt engineering, embeddings, RAG, inference optimisation, evaluation
- Strong Python; production experience with PyTorch, TensorFlow, Hugging Face, LangChain, LlamaIndex, or equivalent
- Track record of deploying and operating AI/ML systems in production
- Solid understanding of data pipelines, model evaluation, monitoring, and MLOps
- Ability to own initiatives independently and communicate clearly with both technical and non-technical stakeholders
- Professional-level English
Nice to have
- Experience with offline or on-device model deployment — edge, mobile, desktop, constrained environments
- Quantisation, distillation, ONNX, TensorFlow Lite, LiteRT, or equivalent optimisation work
- Cybersecurity, privacy, fraud detection, trust & safety, or other security-sensitive AI domains
- End-to-end leadership of LLM initiatives from research/prototype through to production
Why This Role Is Worth Your Time
- LLM architecture ownership is the actual job — you’re not advising or prototyping, you’re building the systems that go into production and defining how the team works in this space
- The cybersecurity domain adds genuine technical constraints: accuracy, safety, and latency requirements that make the engineering harder and more interesting than typical LLM application work
- You’ll shape how a product company adopts AI at a moment when the decisions being made now will be load-bearing for years — that’s a different kind of influence than joining a mature AI org
- On-device and edge deployment is in scope, which is a technically distinct and increasingly relevant problem space — not just cloud inference with a different label
