← All projectsIntermediate · About 10–14 hours
Tool-calling agent with guardrails
An assistant that looks up orders and files tickets, with limits on what it may do on its own.
Good for: AI engineer, Forward deployed engineer
What you will have at the endAn agent that calls 3 tools, asks a human before any write action, stops loops and logs every step.
- Tool calling
- Agent loop and stop conditions
- Human approval
- Input and output validation
- Tracing
New to this? Do the toolkit setup first, then add these one at a time.
- 1
The model chooses tools; your code runs them.
$ uv add openai # or anthropic, or boto3 for Bedrock
- 2
- 3
How to show it to an employer
1Trace screenshot of a full run
2README with a permission tiers table (read, write, never)
3The injection test and how you blocked it
Stretch goals
Prompt-injection test: a document that tells the agent to email data out
Expose the tools as an MCP server
An eval set of 20 tasks with a success rate
Questions this prepares you for
Who executes the tool, the model or your code?
How do you stop an agent looping?
Which actions need human approval and why?
Practise them in a mock interview Next projectMCP server for an internal API→