AI Azure Foundry Engineer – Multi-Agent AI Platform

The employer is a large New Zealand bank building a production-grade, governed multi-agent AI platform on Microsoft Azure AI Foundry. The AI Azure Foundry Engineer will build production agents and multi-agent architectures, integrate enterprise tools and knowledge, and implement identity, security, evaluation, observability, and responsible AI controls. The role also involves collaborating with architecture, risk, compliance, and privacy teams and delivering agents across enterprise and customer-facing channels.

Responsibilities:

• Build conversational and task-oriented agents using Microsoft Foundry Agent Service
• Work with hosted agents, prompt agents and autonomous agent patterns where appropriate
• Develop production-quality solutions using Python and/or C#/.NET
• Configure and manage model deployments, agent capabilities and evaluations within Microsoft Foundry
• Design and implement multi-agent architectures using Microsoft Agent Framework
• Apply supervisor/worker, handoff and Magentic-style orchestration patterns
• Determine when multi-agent orchestration adds genuine value versus when a single well-designed agent and toolset is the better solution
• Design for agent state, memory, context management and failure recovery
• Implement A2A protocol patterns for agent discovery and interoperability
• Work with Agent Cards, capability discovery and delegation
• Enable agents to call remote agents as tools
• Expose enterprise agents for consumption by other agents, platforms and teams
• Build and integrate MCP servers and clients
• Work with Foundry Toolboxes and OpenAPI-based tools
• Design reliable tool interfaces and minimise unnecessary tool complexity
• Implement enterprise knowledge grounding using Foundry IQ and/or Azure AI Search
• Build robust RAG pipelines covering chunking, hybrid search, filtering and reranking
• Diagnose whether poor agent responses are caused by retrieval, context, tools or the underlying model
• Implement agent identity and governance using Microsoft Entra Agent ID and agent blueprints
• Apply Conditional Access and appropriate controls for non-human identities
• Implement OAuth identity passthrough so agents operate within the permissions of the user/customer they represent
• Apply managed identities, Key Vault, RBAC and secure secret management
• Design secure private networking and VNet-integrated AI workloads
• Ensure agents only access data and tools they are authorised to use
• Implement AI evaluation frameworks covering: Groundedness, Relevance, Safety, Accuracy, Tool-use behaviour
• Integrate evaluations into CI/CD pipelines
• Implement tracing and observability using Azure Application Insights and related Azure monitoring capabilities
• Provide sufficient telemetry to understand and audit why an agent produced a particular response or took a particular action
• Implement content filtering and prompt-injection defences
• Design appropriate PII handling and data protection controls
• Build human-in-the-loop checkpoints for high-risk activities, particularly anything involving money, financial decisions or customer advice
• Support red teaming, adversarial testing and AI risk assessment
• Work with risk, compliance, privacy and architecture teams to ensure solutions meet enterprise and regulatory expectations
• Integrate AI agents into enterprise and customer-facing channels including: Microsoft Teams, Microsoft 365 Copilot, Customer-facing web and application experiences