JD - AI Systems Integration Manager (High-Risk AI Compliance)

Section: Position Objective

1. Enable the safe, compliant, and sustainable integration of high-risk and general-purpose AI systems into enterprise architecture, ensuring full alignment with the EU AI Act (Regulation (EU) 2024/1689), harmonised standards (EN 18286, prEN 18228, ISO/IEC 42001), and related legislation (GDPR, NIS-2, DORA, Product Liability Directive).
2. Act as the organisation authoritative technical owner for AI integration strategy, patterns, and standards (SFIA ARCH6, SINT6).
3. Guarantee that every AI integration is modular, observable, traceable, version-controlled, and supports human oversight, rollback, and post-market monitoring (EU AI Act Articles 12, 14, 15, 72).
4. Future-proof integration investments against model drift, supply-chain changes, and evolving regulation.
5. Influence organisational policy and culture on responsible AI integration (SFIA Level 7 behaviour).

Accountable to

Chief AI Governance Officer / Head of AI Architecture (or GM: Architecture if no CAIO exists)

Responsible for (people)

• AI Integration Architects (SFIA SINT5-6)
• MLOps Engineers - Integration Stream (SFIA MLDE5-6)
• Policy-as-Code & Guardrails Engineers
• Data Pipeline Owners touching high-risk systems

Primarily Performs

• Develop and own the Enterprise AI Integration Strategy & Roadmap (SFIA STRT6).
• Define, publish, and enforce AI Integration Reference Architecture & Pattern Library (approved RAG, agentic, tool-calling, and multi-modal patterns) (SFIA ARCH6, SINT6).
• Lead the AI Integration Architecture Review Board with veto rights on non-conforming designs (SFIA Level 6 autonomy).
• Maintain living artefacts: AI Integration Catalogue, SBOMs, model + prompt + embedding lineage maps, system cards (SFIA DTAN6, ARCD6).
• Perform technical studies and proof-of-concepts on emerging integration technologies (LangChain, LlamaIndex, DSPy, AutoGen, Vertex AI Agent Builder, Azure AI Studio, etc.) (SFIA TECH6).
• Implement and continuously improve policy-as-code guardrails (Open Policy Agent, Conftest, custom AI semantic conventions) for prompt injection, personal data leakage, tool-calling safety, and content moderation.
• Conduct regular audits and red-team testing of production AI integrations (SFIA TEST6, SECM6).
• Deliver mandatory training and awareness programmes on AI integration standards and regulatory obligations (SFIA EVLN6).
• Act as single technical point of contact for notified bodies during conformity assessments of integration layers.
• Conduct regular audits of production integrations for conformance to AI integration standards and regulatory obligations.
• Review the AI integration design of all key projects (especially Annex III high-risk systems) and sign off before go-live.

Key Deliverables

• Annual Enterprise AI Integration Strategy & Roadmap (signed off at management review)
• AI Integration Reference Architecture (Confluence/Notion living site) & Pattern Library
• Quarterly AI Integration Compliance & Performance Report (presented to management review - SFIA METL6)
• Architecture Decision Records (ADRs) for every new pattern or tool
• Automated compliance gate scan reports (OPA + AI-specific policies)
• Full lineage graphs and SBOMs for every high-risk integration

Concurs / Collaborates with

• AI Governance Specialist & Compliance Officer
• Chief Data Officer / Data Architects
• Security Architect (cyber & adversarial robustness)
• MLOps & Platform Engineering Leads
• Risk Management Lead (Article 9)
• Human Oversight & Safety Team
• External Notified Body / Regulators (when required)

Key Result Areas

• 100% of high-risk AI integrations reviewed and approved before production (target > 95% first-pass approval)
• Zero major or critical findings related to integration in notified-body or market-surveillance audits
• Mean time to trace a production output to exact prompt + retrieved chunks + model version < 5 minutes
• Prompt-injection / jailbreak success rate in red-team tests < 1%
• All production integrations pass automated compliance gate on every deploy
• Stakeholder satisfaction (compliance, security, data science) = 4.5/5

Knowledge Required

• EU AI Act (full text) + harmonised standards (EN 18286, prEN 18228)
• SFIA Level 6 proficiency in SINT, ARCH, MLDE, GOVC, RISM
• Modern AI integration stacks (LangChain/LangGraph, LlamaIndex, DSPy, AutoGen, Vertex AI Agent Builder, Azure AI Studio)
• Observability for LLM applications (LangSmith, Phoenix, Helicone, OpenTelemetry AI extensions)
• Policy-as-code (OPA, Gatekeeper) and guardrails
• Model cards, system cards, SBOMs, lineage standards (OpenLineage + AI-specific)

Core Skills Required

• Architecture decision-making under regulatory constraints
• Policy-as-code and automated guardrails (Open Policy Agent, Gatekeeper)
• Communicating complex AI integration risks to executives and regulators
• Running effective architecture review boards
• Incident triage and root-cause analysis for integration failures affecting high-risk systems

Personal Attributes

• Regulatory mind-set combined with technical depth
• Assertive enforcement of standards without blocking innovation
• Strong stakeholder management across legal, risk, and engineering

Qualifications & Experience

• Degree in Computer Science, AI Engineering, or equivalent
• Minimum 2 years in enterprise integration / middleware / API management
• Minimum 2 years hands-on with production LLM / generative AI integrations
• Demonstrate In-depth Understanding (any combination): EN 18286, prEN 18228, ISO/IEC 42001, ITIL 4, Cloud AI (Google Professional ML Engineer, Azure AI Engineer, AWS ML Specialty), COBIT, TOGAF 9/10 or equivalent.
• Proven track record of delivering high-risk or safety-critical integrations in regulated environments (finance, health, aviation, or public sector preferred)