Guardrails in Practice
Guardrails are the checks around an AI feature that stop it from saying or doing the wrong thing. You will add them to Ask Corri, a made-up insurer's help assistant, in real code: a check on what comes in, a check on what goes out, rules taken from a one-page AI use policy, and a step where a person approves every refund. Then you measure what each guardrail catches and what it wrongly blocks, and see that a stricter check catches more but stops more real customers. Runs with no API key; all practice data is made up. Educational, not legal advice.
"Every risky thing Ask Corri could do has a guardrail, we know what each one catches and wrongly blocks, and risky actions wait for a person"
6 Interactive Sessions
Short, interactive sessions: watch it work, try it, then build it yourself. You can go deeper at any time in the written chapters.
- 1
What a guardrail is, and where it sits
A guardrail is a check around an AI feature that stops it saying or doing the wrong thing. You find every place in the flow where one can sit, and mark it.
- 2
Checking what comes in
An input check reads each customer message before the model does. You score it on 200 labelled messages, and see that a stricter check catches more problems and blocks more good customers.
- 3
Checking what goes out
An output check reads each answer before the customer does. You build one that catches personal data, made-up cover names and promises about claims, and you find what it still misses.
- 4
Turning a policy into rules
A policy written for people only protects you when each line becomes something you can test. You sort Corriwell's ten policy lines, turn seven into code rules with test cases, and fix the one that fails.
- 5
A person in the loop
Some actions cost money or cannot be undone. Ask Corri may ask for them, but a person decides. You build the approval step for request_refund and design what the approver sees.
- 6
The cost of guardrails
Every guardrail catches some problems, stops some good things by mistake, and adds time. You put all four in one report, check a summary of it, and decide what to change next.
Skills you'll practise
Practice data included
- 200 labelled customer messages
- 30 draft answers with planted problems
- A one-page AI use policy with rule test cases
- 12 claims and 16 refund requests
What you walk away with
Shareable portfolio
A public URL showing your module timeline, patterns mastered, and completion status.
Everything you build
Download all your work as a ZIP file. It is yours to keep.
Module walkthrough
Each module documented with deliverables and the production pattern you implemented.
Ready to start Guardrails in Practice?
First course free. $20 per course after that.