Your cart is empty.
Keep shopping
AI Architecture: Designing LLM Systems for the Enterprise
New
11 students
AI Architecture: Designing LLM Systems for the Enterprise
AI architecture patterns for the enterprise: RAG, agentic workflows, agents, MCP, security and governance
Created byDr. Florian Detzel | BI Academy
Last updated 10/2026
English
What you'll learn
- Choose the right AI architecture pattern, from a single model call to multi-agent systems, and document it with C4 diagrams.
- Assess whether an AI use case is feasible and worth building, and record the decision in an architecture decision record.
- Design a model gateway that controls which models are used, where data is processed and what each call costs.
- Design RAG that finds the right document version and respects access rights, and decide when prompting or fine-tuning fits better.
- Design AI agents with clear limits on what they may see, do and spend, and decide when MCP or a second agent is worth it.
- Secure and govern LLM systems: guardrails against prompt injection, data residency, EU AI Act classification and audit trails.
- Evaluate an AI system with baselines and release gates, and review a complete AI architecture before it goes live.
Explore related topics
Course content
14 sections • 121 lectures • 11h 42m total length
Requirements
- Comfortable reading technical documents: a log, a configuration file, a JSON record. You read them; you do not write them.
- No AI or machine learning background is assumed. No GPU, Docker, virtual machine, IDE, Jupyter or database server is needed. Any laptop, any operating system.
Description
This course teaches you to design the architecture of AI systems that run inside a real company: everything around the model that decides whether it is safe, affordable and allowed to go live.
You work on one case study. Halden Insurance is a fictional European insurer with old core systems, customers in two countries and a compliance team that reads everything. Module by module, we discuss AI architecture: We start with fundamentals, move over to a gateway in front of the model, retrieval that only shows people the documents they are allowed to see, an agent with a budget and a clear way to stop, protection against prompt injection, an audit trail, the EU AI Act classification, and the evaluation that decides whether a release goes out. In the last module you review the whole system the way a design board would and work out what still needs fixing.
Eleven of the twelve modules end with a hands-on lab. You write the documents architects actually get asked for, like a decision record, an agent charter, a threat model or a go-live pack, and you draw the architecture yourself as it grows. Every lab comes with a worked solution, and every module with a short quiz.
You don't need to code.
You will learn to think like an AI Architect. If you're an architect or senior engineer who has just been handed "the AI project", this course is for you.
Note for transparency: Some content of this course has been created with the help of AI. This course is no AI slop!
Who this course is for:
- Software, solution and enterprise architects who have been handed an AI initiative
- Senior engineers and tech leads making the decisions that an architect would normally make.
- Anyone who already owns a system that an AI component now has to enter, which is the situation the whole course is set in.
- Architects who want the governance, security and evaluation half taken as seriously as the pattern half.