AI Solution Architect

Learn the whole job of an AI solution architect: understand what a business actually needs, decide whether AI should be used at all, choose the simplest pattern that works, price it, design it so it survives production, security review and regulation — and explain every decision to an executive and to an engineer in the same afternoon.

An AI engineer is asked “how do I build this?”. An AI solution architect is asked the three questions that come before it — should we build this, what exactly should we build, and what will it cost to run for three years? — and then the one that comes after: why did it not reach production?

This course starts with someone who has never called a model, never used a cloud console and has never been in a customer meeting. It ends with a complete engagement worked from the first discovery call to the steering-committee decision, with the numbers shown.

It assumes you can read simple code. It does not assume you know machine learning, cloud, or how enterprises buy software.

What the job produces#

flowchart LR
  P[":i-building-2: <b>A business problem</b><br/><small>slow, costly or impossible today</small>"] --> D[":i-search: <b>Discovery</b><br/><small>who, what, how often, what data</small>"]
  D --> Q[":i-funnel: <b>Qualify</b><br/><small>value, feasibility, risk</small>"]
  Q -->|"not worth it"| N[":i-ban: <b>A clear no</b><br/><small>also a deliverable</small>"]
  Q --> S[":i-layers: <b>Solution design</b><br/><small>pattern, model, platform</small>"]
  S --> C[":i-coins: <b>Cost and ROI</b><br/><small>per task, per year</small>"]
  C --> G[":i-shield-check: <b>Risk and governance</b><br/><small>security, privacy, law</small>"]
  G --> X[":i-file-text: <b>Decision document</b><br/><small>design, ADRs, plan</small>"]
  X --> R[":i-activity: <b>Pilot to production</b><br/><small>evals gate every stage</small>"]
  class P,N neutral
  class D,Q queue
  class S compute
  class C memory
  class G warn
  class X,R io

Every box in that picture is a lesson. The arrow labelled “not worth it” matters as much as the others: an architect who can say no early, with a reason, saves more money than one who designs well.

How this course works#

Every lesson follows the same shape:

1. The idea in one minute — the whole lesson in a few sentences.
2. A picture — the mechanism or the decision, with real tools drawn in where they matter. Press Expand on any diagram to open it full size.
3. How it really works — the precise version, with names, numbers and templates you can copy.
4. Code — where arithmetic drives a decision, a small Go program that uses only the standard library. Architects do not need to be programmers, but they do need to rerun the numbers when an assumption changes.
5. Remember this, Try it and Check yourself.

Three rules run through every lesson:

• Problem first, pattern second, product last. A tool named in a lesson is an example of what fills a box, not an endorsement.
• The simplest thing that passes the evaluation wins. Complexity has to be paid for with evidence.
• Every recommendation carries a number and a way to be proved wrong.

The topics#

flowchart LR
  A["The role"] --> B["The solution<br/>toolbox"]
  B --> C["From problem<br/>to design"]
  C --> D["Production<br/>architecture"]
  D --> E["Enterprise<br/>strategy"]
  E --> F["The career"]
  class A neutral
  class B compute
  class C queue
  class D warn
  class E memory
  class F io

Two reference pages sit beside the topics: the Glossary, and The Best of the Internet — the short list of free guides, frameworks, books and courses worth your time, each with a line on what it is good for.

How it connects to the other courses#

This course is about deciding and justifying. When a decision needs depth, another course has it:

• How to draw the system itself, box by box: AI System Design.
• What happens when you host models yourself: Inference Engineering and vLLM.
• What the hardware is and costs: GPU Engineering.
• How to see it running: Observability Engineering.
• How it is attacked and defended: AI Security Engineering.

You can finish this course without reading any of them. The links are there for the day a customer asks the follow-up question.

What you need#

• Go 1.22 or newer and a terminal, to run the calculators. Every program runs offline.
• No GPU, no cloud account and no API key.
• A notebook. Half of this job is writing things down so that other people can disagree with them.

A promise about names and numbers#

Prices, product names, certification details and legal dates in this course were checked on 13 October 2026, and each lesson that depends on them lists its sources. Product names in this field change about once a year — two of the three large cloud AI platforms were renamed in 2026 alone. The method — discover, qualify, design, price, de-risk, decide — does not change, and every lesson keeps the two apart so you know which half to re-check.