AI Engineer

This role gets AI into production and keeps it running. You own the MLOps backbone and build the agentic systems we deliver for clients: agents, GraphRAG, enterprise copilots, and workflow automation. The particular challenge here is dependability. An LLM system that dazzles in a demo can behave unpredictably in a client's real workflow, and your job is to make it steady: reliable, affordable to run, guarded, and watched. That's the problem this role turns over, and if it's the kind of problem you enjoy, there's plenty of it. **The work:** * Design and ship agentic systems: agents, RAG and GraphRAG, copilots the client's people actually use * Own the model-specific side of production: serving logic, evaluation, guardrails, drift and behavior monitoring, and the retraining loop * Build the MLOps frameworks and reusable AI infrastructure every engagement draws on * Expose models and agents behind clean endpoints for the applications to consume * Integrate AI into the client's real operational workflows, not tools that sit unused **Where your work ends.** You take models from the Data Scientist and the graph foundation from the Data Engineer and turn them into dependable production AI. You own everything model-specific, but you run it on the platform the Software Engineer provides: you don't own the containers, Kubernetes, cloud provisioning, or the generic observability stack, and you don't build the application front ends. Your line is model behavior; theirs is the platform and the product surface. **What success looks like:** * AI systems stay reliable in a client's real workflow, not just in a demo * Guardrails, evals, and monitoring are in place before anything goes live * The cost of running AI stays predictable and defensible * The AI infrastructure you build makes the next engagement faster, not slower The stack we work in today: Python for the model and service work, with FastAPI where you're exposing a model or agent behind an endpoint; PyTorch, with real fluency in LLM internals, fine-tuning, and, where it earns its place, mechanistic interpretability; MCP and A2A for agent and tool interop; and Neo4j behind our GraphRAG systems. A depth in a serious slice of this, plus the judgment to learn the rest, matters more than checking every box. For this role, MLOps is the craft itself. **About The Strong AI, and how we work** The Strong AI is an end-to-end AI implementation partner. Clients come to us because most organizations can run an AI experiment, but few can turn it into a system their business depends on. We close that gap. We don't hand over slideware or a notebook; we build systems that work inside a client's business, and where they want it, we run them. You'll work across engagements and industries, on different problems and often different stacks. We're technology-agnostic: the problem and the client's environment choose the tools, so treat any stack we list as the ground we work on today, not a gate. **Across all roles, we ask for the same way of working:** * Real software. Tested, reviewed, versioned code the next person, or the client's team, can pick up. * MLOps mindset. A model's life starts at deployment. Monitoring, retraining, drift, and rollback are handled before anything breaks. * Systems thinking. You see both the value slice and the whole it compounds into. * Quality and security, owned by you. Designed in from the first decision, not inspected in at the end. Everyone builds to the highest standard. * Built for handover. Clear code and docs the client's own team can understand, operate, and take over.