- Lab
-
Libraries: If you want this lab, consider one of these libraries.
- Cloud
Gate a Google Cloud GenAI Workflow for Limited Release
A customer care team wants its Gemini reply workflow to start answering some customers on its own. The pilot demo was convincing, but nobody has checked the workflow against the company's limited-release standard. In this lab, you will assess the workflow on Google Cloud against that standard, rank its gaps by the harm each could cause, close them by configuring Model Armor screening and a release scope instead of rewriting the prompt, test representative messages to prove each safeguard engages, and record what the workflow is cleared to do and which cases still go to a person.
Lab Info
Table of Contents
-
Challenge
Assess the workflow against the readiness standard and identify unmet requirements
- Review the limited-release standard and the workflow's current settings.
- Review the pilot results for evidence on each requirement.
- Record which requirements the workflow meets and which it does not.
At the end of this objective, you'll know which requirements Reply Assist already meets and which ones stand between it and a limited release.
-
Challenge
Prioritize the gaps by the potential harm if each one is left unresolved
- Rate each unmet requirement by who could be harmed, how badly, and whether the harm can be undone.
- Order the gaps from most to least harmful.
- Record the order in the release record.
At the end of this objective, you'll have a ranked list of gaps that sets the order in which you close them.
-
Challenge
Configure safeguards to close the prioritized gaps without changing the prompt
- Turn on the Model Armor filters the standard calls for.
- Switch the workflow to enforce screening and limit automatic sending to the cleared scope.
- Deploy the change and confirm that the new revision is live.
At the end of this objective, you'll have a new revision of Reply Assist live with every prioritized safeguard switched on and the prompt unchanged.
-
Challenge
Test representative cases and verify that each safeguard engages as intended
- Submit the six representative test messages.
- Check where each message went and which check decided its route.
At the end of this objective, you'll have seen each safeguard engage on the case it is meant to catch and the plain order-status question go out automatically.
-
Challenge
Document what the workflow is cleared to do and which cases still require human review
- Record the cleared scope and the safeguards you verified for it.
- Record which cases still require human review and why.
At the end of this objective, you'll have a release record that states what Reply Assist may do on its own and what still goes to a person.
Real skill practice before real-world application
Hands-on Labs are real environments created by industry experts to help you learn. These environments help you gain knowledge and experience, practice without compromising your system, test without risk, destroy without fear, and let you learn from your mistakes. Hands-on Labs: practice your skills before delivering in the real world.
Learn by doing
Engage hands-on with the tools and technologies you’re learning. You pick the skill, we provide the credentials and environment.
Follow your guide
All labs have detailed instructions and objectives, guiding you through the learning process and ensuring you understand every step.
Turn time into mastery
On average, you retain 75% more of your learning if you take time to practice. Hands-on labs set you up for success to make those skills stick.