#EG AI Engineer (Autonomous Network) at NCS3

The AI Engineer builds, tests, and iterates production Agentic AI systems at the core of the Autonomous Networks Programme. Working directly with the AI Solution Architect, Business Analyst, and Central Infrastructure Solution Architect, this role turns architecture designs into working code — implementing agent logic, integrating OSS APIs, building RAG pipelines, and deploying agents into staging and production environments. This is a hands-on engineering and development role where code quality, test rigour, and operational reliability are the primary measures of success.   What will you do? Technical Delivery • Own end-to-end delivery of AI workstreams: from requirements through data preparation, modelling, integration, testing, and production handover • Develop agentic applications including RAG pipelines, prompt-engineered agents, and agentic workflows using LangChain, LlamaIndex, LangGraph, or plain Python • Build on top of GenAI application stacks including LLM orchestration, observability (Langfuse, Braintrust), guardrails, and LLM gateway patterns (LiteLLM, Portkey) • Implement multi-agent orchestration layer: event routing, resource locking, inter-agent handoff contracts, prompt caching, and shared state management • Implement agentic design patterns including workflow evaluation, LLM-as-Judge, and AI red teaming • Adopt Model Context Protocol (MCP) and Agent-to-Agent (A2A) Protocol as foundational extension mechanisms, enabling agents to operate safely across enterprise ecosystems • Work with AI Solution Architect and Business Analysts on benchmarking and golden-set compilation for agent evaluation • Implement effective RAG architectures: chunking strategies, embedding selection, vector store configuration (Qdrant, Milvus, pgvector), hybrid search, and reranking • Build and maintain evaluation harnesses measuring correctness, faithfulness, latency, and safety. Run continuous LLM-as-Judge evaluation on production traces • Write clean, maintainable Python; participate in code reviews; contribute to shared AI platform librariesOSS Integration & Data Engineering • Implement REST, NETCONF/YANG, SNMP trap ingestion, and gNMI streaming telemetry integrations connecting agents to NMS and EMS systems • Build ServiceNow integrations for ticket read/write, triage updates, change record queries, and human-in-the-loop approval workflow triggers for the Execution Agent • Build the alarm and telemetry data normalisation pipeline converting multi-vendor OSS data into agent-consumable schemas • Implement Kafka or equivalent event streaming consumers delivering real-time alarm data to the Alarm Correlation and Ticket Triaging agents • Work with data and integration engineers to normalise multi-vendor, multi-format network telemetry into consistent schemas the agents can reason overSafety, Guardrails & Risk • Implement guardrail architectures: hard limits, soft limits with human confirmation flows, blast radius controls, and escalation logic • Assist the AI Solution Architect to assess blast radius for each agent — documenting worst-case impacts and designing specific mitigating controls • Implement and validate rollback procedures for all agent-initiated network actions, including automated rollback triggers based on post-execution KPI degradation • Work with QA engineers to design and test solutions via defined test cases, and review required improvements • Ensure all agent architectures comply with IMDA regulatory requirements, AIVerify framework requirements, InfoSec policies, and CMB change governance frameworksObservability & Evaluation • Implement the Decision Audit Trail schema for all agents ensuring every decision is fully traceable, explainable, and available for post-incident review • Build the Agent Operations Dashboard for the NOC team: real-time agent status, decision volumes, accuracy metrics, and exception alerts • Define and implement the evaluation framework: factual, reasoning, tool-call, retrieval and planning accuracy, confidence thresholds, and human feedback capture • Implement continuous evaluation of production agent traces using LLM-as-Judge for quality and safety; design human evaluation sample process • Build drift detection mechanisms that alert the team when agent output quality degrades before it becomes an operational incident • Containerise and deploy agent services using Docker and CI/CD pipelines; manage staging and production deployments with rollback capability