Agentic Artificial Intelligence (AI) Engineer

Location: Tukang, Singapore (West) Only candidates with the legal right to work in Singapore will be considered. Please note that applicants without this eligibility will not be considered. Why join Rolls-Royce? At Rolls-Royce we are proud to be a business that has truly helped to shape the modern world and are committed to always being a force for progress; powering, protecting and connecting people everywhere. By joining Rolls-Royce, you'll have the opportunity to work on world-class solutions, supported by a culture that believes individuality is our greatest strength, and all perspectives, experiences and backgrounds help us innovate and enable our high-performance culture. Position Summary The Agentic AI Engineer designs and delivers the agents that power RRPS's AI-driven operations. This role is accountable for applying LLM-first design principles: every agent built here uses the language model as its reasoning core, with tools and APIs as stateless data endpoints. Engineers in this role maintain rigorous audit trails so every autonomous decision is traceable to a named business authority, and they deliver complete working agents — not prototypes — within few weeks. What you will be doing: LLM-First Agent Design & Delivery Design autonomous agents where the LLM performs all reasoning, classification, and routing — no if-else decision logic in agent code. Implement agent tool interfaces as stateless, deterministic data endpoints; all business logic lives in the model's reasoning layer. Deliver production-ready agents within 4–6 week cycles, from capability design through staging validation to live deployment. Apply structured output schemas and constraint enforcement to ensure agents produce auditable, machine-verifiable responses. Write comprehensive agent test suites covering happy paths, adversarial inputs, and boundary-condition reasoning. Multi-Agent Orchestration & Graduated Autonomy Design multi-agent systems where specialist agents collaborate under a coordinating orchestrator, with clearly bounded authority at each layer. Implement graduated autonomy protocols: move agents from supervised pilot through validated autonomous to fully autonomous deployment, with defined quality gates at each stage. Enforce the agent's sanctioned authority boundaries at runtime — reject tool calls that exceed what the agent is permitted to do, before execution. Build agent-to-agent communication patterns that preserve audit context across handoffs, ensuring end-to-end traceability. Monitor agent behaviour in production; detect drift from approved operating parameters and escalate to human oversight. Audit, Compliance & Institutional Knowledge Implement comprehensive audit trails capturing decision rationale, tool calls, inputs, outputs, and the human authority chain behind each action. Ensure every agent output includes sufficient explanation for regulatory review under EU AI Act transparency obligations. Document agent design decisions, authority boundary definitions, and known limitations in the CoE knowledge base. Support internal compliance authorities consultations by providing technical documentation of agent capabilities and human oversight controls. Contribute to the CoE's institutional knowledge library — patterns, lessons learned, and reusable agent components. Platform Integration & Quality Integrate agents with enterprise systems via standard API contracts, event queues, and Model Context Protocol (MCP) tool interfaces. Implement idempotent agent actions — every autonomous operation must be safe to retry without side effects. Apply structured validation at all agent output boundaries before downstream system calls. Participate in CoE code reviews, contributing to shared standards for agent architecture and prompt engineering. Maintain agent health monitoring: latency, reliability, reasoning quality metrics, and escalation rates. Position requirements: LLM-first design: understanding that the model IS the router, classifier, and evaluator — not a call made by deterministic code. Agent reliability patterns: idempotency, retry safety, graceful degradation without silent fallback. AI audit and explainability: designing for regulatory review, not just functional correctness. EU AI Act high-risk system obligations (basic awareness sufficient). Enterprise integration patterns: event-driven architecture, saga pattern, API gateway, service mesh. Technical Skills Required LLM integration: Anthropic Claude, Azure OpenAI…