Architecting AI Systems

About this course

The flagship course of the AI Solution Architecture track. Learners architect production AI systems the way an enterprise review board expects: from framing the use case and choosing models and adaptation patterns, through retrieval, agent and integration architecture, to the evaluation, observability, security, cost and governance layers that make an AI system operable and compliant. Every decision is recorded in ADRs and defended in a live review.

For: architect, tech leader, ml practitioner, software engineer.

Learning outcomes

On successful completion you will be able to:

1. 1Explain the reference architecture of enterprise AI systems (experience, orchestration, model, retrieval, data, gateway, evaluation and governance layers) and how each affects the system's quality attributes.
2. 2Analyse an AI use case for value, feasibility, risk and data readiness, and express it as architecturally significant requirements with measurable targets for quality, latency, cost and safety.
3. 3Evaluate model and pattern options (prompting, retrieval, fine-tuning, agents; hosted or open-weight; build or buy) against explicit criteria and record the decision in an architecture decision record.
4. 4Design an end-to-end AI solution architecture with C4 views, data flows, failure modes, an evaluation strategy and a cost model.
5. 5Design security, privacy and safety controls for an LLM application, including prompt-injection defence, guardrails, data protection and human oversight.
6. 6Lead an architecture review of an AI system, defending trade-offs and residual risk to technical and executive stakeholders and taking responsibility for GDPR and EU AI Act obligations.
7. 7Define the operating model of an AI system (LLMOps, ownership, release gates, incident response and model lifecycle) under uncertainty about fast-moving technology.

Modules and lessons

1. Week 1Framing AI solutions⌄

   • The reference architecture of enterprise AI systems
   • From use case to architecturally significant requirements
2. Week 2Choosing models and patterns⌄

   • Model selection: hosted, open-weight and small models
   • Prompting, retrieval, fine-tuning or agents: making the calllab
3. Week 3Retrieval and knowledge architecture⌄

   • Retrieval-augmented generation at enterprise scale
   • Structured data, knowledge graphs and text-to-SQLcase study
4. Week 4Agentic and integration architecture⌄

   • Agent architectures for the enterprise
   • Integrating AI into existing systemslab
5. Week 5Quality: evaluation and observability⌄

   • Evaluation as an architectural concern
   • Observability for AI systems
6. Week 6Security, safety and trust⌄

   • Threat modelling LLM applicationslab
   • Guardrails, privacy and human oversight
7. Week 7Cost, scale and operations⌄

   • Cost architecture and FinOps for AIlab
   • LLMOps and the operating model
8. Week 8Governance and the review board⌄

   • Governing AI architecture
   • Defending an AI architecture at the review boardcase study

Assessment

• Quiz (15%)

  Eight weekly quizzes drawn from the lesson quiz banks; best seven count.

• Design review (35%)

  Architecture package for a provided enterprise case - C4 context and container views, three ADRs, an evaluation plan, a threat model and a cost model.

• Project (25%)

  Use-case canvas and operating model for an AI capability in the learner's own organisation, or a provided case if confidentiality prevents it.

• Viva (25%)

  Thirty-minute review-board defence of the design package with two mentors acting as CISO and CFO.

Pass mark 60%. Grades: Distinction, Merit, Pass.