AI Solution Architect Course for Enterprise AI Architecture

01AI Opportunity Discovery and Business Outcomes

• ·Map the current workflow, business capabilities, stakeholders and decision points before proposing AI.
• ·Compare conventional software, search, rules, analytics, ML and generative AI against the actual problem.
• ·Define measurable outcomes, adoption constraints and the cost of an incorrect decision or action.

Design output: An opportunity brief explaining whether AI is justified and how value would be measured.

02Requirements and Quality Attributes

• ·Translate stakeholder expectations into functional requirements and measurable non-functional requirements (NFRs).
• ·Define latency, throughput, availability, privacy, security, accessibility and recovery expectations.
• ·Resolve competing requirements, including data residency, auditability, operating cost and delivery time.

Design output: A requirements and acceptance matrix with assumptions, priorities and accountable owners.

03Enterprise Context and Target Architecture

• ·Map the current application, data, identity, network and integration landscape.
• ·Define solution boundaries and a target state that fits the existing estate.
• ·Identify gaps, dependencies and transition stages without turning enterprise architecture into a notation exercise.

Design output: Current-state and target-state diagrams, with the changes needed to move between them.

04AI Architecture Decisions

• ·Compare retrieval, model adaptation and tool use by the capability each adds. RAG and fine-tuning may be used together.
• ·Choose workflow or agent behavior, then test whether multiple agents solve a specific coordination need.
• ·Compare hosted and self-hosted models, managed services and custom platforms using constraints rather than a preferred stack.

Design output: A decision matrix and architecture decision records (ADRs) that make alternatives and rejected options explicit.

05Enterprise Data Architecture and Governance

• ·Map structured and unstructured data, ownership, contracts and access boundaries.
• ·Compare batch, incremental and streaming integration using freshness and operational requirements.
• ·Define lineage, quality, retention, deletion, residency and reference-data responsibilities across the solution.

Design output: A data-flow and governance map showing who owns the data and where access and lifecycle rules apply.

06Generative AI and RAG Solution Architecture

• ·Design enterprise knowledge workflows with ingestion, retrieval, reranking, model access and answer evaluation.
• ·Connect identity and document permissions to retrieval, caching and evidence shown to the user.
• ·Define quality acceptance, human review and fallback behavior alongside prompt and configuration management.

Design output: A knowledge-solution design with clear ownership for relevance, access, freshness and answer quality.

For implementation depth in RAG, model adaptation and evaluation, explore the Generative AI Course .

07Agentic Enterprise Architecture

• ·Place agents inside business workflows with explicit boundaries between recommendations, approvals and execution.
• ·Define tool identities, permissions, state, memory, durable execution and audit records across enterprise integrations.
• ·Evaluate MCP-based integration, agent coordination and shared runtime controls without assuming that a protocol supplies authorization or reliable execution by itself.

Design output: An agent operating design showing permitted actions, approval ownership, execution limits and recovery responsibilities.

For hands-on orchestration, tool controls and stateful agents, explore the Agentic AI Course .

08Cloud, Platform and Enterprise Integration

• ·Compare AWS, Azure and Google Cloud through model access, custom deployment, enterprise identity, networking, data integration and operational ownership.
• ·Map patterns to representative services, including Amazon Bedrock, Amazon SageMaker AI and Microsoft Foundry. Compare capabilities and constraints; do not assume that similarly positioned services are interchangeable.
• ·Evaluate managed APIs, containers, Kubernetes, hybrid connections, event integration and API gateways where the requirements justify them.

Design output: A platform recommendation with integration boundaries, constraints and a credible operating model. The work focuses on architecture choices rather than console walkthroughs for every cloud.

For deeper ML lifecycle workflows, explore the MLOps Course . For LLM releases, serving and operating workflows, explore the LLMOps Course .

09AI Security, Governance and Risk

• ·Threat-model prompt injection, data exposure, unsafe tool actions, poisoned content and vendor dependencies.
• ·Connect each material risk to controls, owners, test evidence and escalation paths. Combine least privilege, isolation, validation and human oversight.
• ·Define AI inventory, release approval, audit evidence and ongoing review. Treat legal and regulatory requirements as inputs requiring qualified interpretation.

Design output: A risk register and control map that shows what is prevented, detected, reviewed and still unresolved.

10AI Economics, FinOps and Build-versus-Buy

• ·Estimate model usage, hosting, search, storage, network, licensing, observability and support costs.
• ·Include integration effort, evaluation, human review, redundancy and ongoing engineering work in total cost of ownership (TCO).
• ·Compare cost per accepted outcome, not just token price or nominal request cost. Test sensitivity to traffic, retry rates, review effort and utilization.

Design output: A build-versus-buy recommendation with a cost model, comparison horizon and assumptions that can be updated.

11Architecture Communication and Decision Records

• ·Use system-context, component, sequence, data-flow and deployment diagrams to answer specific stakeholder questions.
• ·Document quality attributes, alternatives, risks and decisions in a form that engineers can implement and reviewers can challenge.
• ·Present a technical proposal and an executive summary without losing important constraints in either version.

Design output: An architecture pack that connects business outcomes to technical choices and unresolved decisions.

12Architecture Review, Rollout and Ownership

• ·Review the solution against its requirements, quality evidence, security controls, cost assumptions and delivery dependencies.
• ·Define a pilot, rollout gates, acceptance criteria, rollback triggers and a path for adoption.
• ·Assign ownership for models, data, prompts, tools, evaluation, incidents and changes after launch.

Design output: A phased delivery and operating plan that explains how the organization will adopt, monitor and revise the solution.

Enterprise Knowledge Across Different Access Boundaries

Business requirement

Employees need answers across HR, support and engineering content without exposing restricted material.

Options

Extend the existing enterprise search platform, use a managed knowledge service or build a custom retrieval application.

Decisions to defend

Data ownership, identity integration, permission enforcement, freshness, evaluation and which parts are worth building internally.

Risk and economics

Connectors, support work and authorization changes may dominate the effort even when model usage is inexpensive.

Review evidence

A target architecture, permission and deletion tests, an evaluation plan and a cost comparison for the stated workload.

Customer Service with Limited Automation

Business requirement

Improve support handling while keeping sensitive actions under appropriate human control.

Options

Agent assistance, a constrained workflow or an agent that can initiate approved business actions.

Decisions to defend

Which actions are permitted, how identity reaches downstream systems, when approval is required and how failed or disputed actions are recovered.

Risk and economics

Compare cost per resolved case, escalation effort, error handling and the integration cost of the existing support platform.

Review evidence

An action-permission matrix, workflow design, audit plan and staged rollout criteria.

A Coding Assistant for Private Repositories

Business requirement

Help engineering teams work with private code while respecting repository access, review and software supply-chain controls.

Options

An existing coding-assistant product, a managed enterprise platform or a custom solution for a narrow internal workflow.

Decisions to defend

Context access, model and vendor policy, tool execution, generated-code review and the boundary between suggestions and changes.

Risk and economics

Balance licensing and integration effort against support ownership, developer adoption and exposure of code or secrets.

Review evidence

A build-versus-buy matrix, data-flow diagram, tool policy and a pilot evaluation based on agreed engineering outcomes.

Document Intelligence with an Auditable Review Path

Business requirement

Extract information from business documents and route uncertain results to the people responsible for review.

Options

A specialist document service, a model-based extraction pipeline or a hybrid workflow with conventional parsing and human checks.

Decisions to defend

Data residency, evidence retention, validation, exception handling and integration with the system of record.

Risk and economics

Include manual review, correction, reprocessing and retention costs rather than comparing extraction API prices alone.

Review evidence

A target design, control map, review workflow and cost model tied to accepted records.

What Your Architecture Pack Contains

• ·Business problem, stakeholder needs and measurable success criteria.
• ·Functional requirements, NFRs and the decision to use AI or a conventional alternative.
• ·Current-state context, solution alternatives and target architecture.
• ·Application, data, AI, integration, deployment and security views.
• ·A risk register, governance controls and assigned operating responsibilities.
• ·Build-versus-buy analysis and TCO assumptions, including human work.
• ·A pilot, phased rollout, acceptance criteria and rollback triggers.
• ·ADRs, a technical proposal and an executive architecture summary.

What the Review Examines

You defend the recommendation and revise it when the evidence does not support an assumption. The goal is an explainable solution decision, supported by architecture and delivery evidence.