Final Project: Build an AI Assistant

Build an assistant that solves one defined problem. Write a short specification, create a structured output contract, connect one safe tool, add tests and an evaluation set, and document how a human reviews the result before it is used.

Knowledge check

AI Engineering Final Assessment

Review the full course before starting your project.

1. What is the application’s role in an AI system?

Validate output, enforce permissions, and own stateGuess what the user wants without checksReplace all testsExpose secrets to the model

2. What is the first step for a new AI feature?

Define the problem and output contractChoose random toolsDeploy immediatelySkip evaluation

3. Why use retrieval?

To provide relevant, current source contextTo make a model consciousTo remove all dataTo avoid citations

4. Why are structured outputs valuable?

They make results easier to parse and validateThey guarantee truthThey remove latencyThey replace a database

5. What should happen before a consequential tool action?

Validate authorization and, when appropriate, request approvalExecute automaticallyPrint the API keyIgnore the user

6. What should tests measure?

Contracts, safety, failure behavior, and qualityOnly one demo responseColor contrast onlyNothing after launch

7. What does a cost budget help with?

Keeping usage predictable and observableMaking prompts longerRemoving securityDisabling logs

8. What is a safe deployment practice?

Use secrets management, timeouts, monitoring, and rollbackCommit secretsSkip health checksDeploy without a plan

9. Why keep a human review path?

People remain accountable for consequential outcomesModels cannot produce draftsIt removes all automationIt avoids writing specifications

10. What makes an AI feature production-ready?

A tested workflow that is useful, observable, and controlledA clever prompt aloneA large model aloneA hidden endpoint

← Previous

Making sense of the next wave.

About Topics Courses