Study StudioCertificatesDashboardSign inSign in

← All projects

Capstoneadvanced8h

AI Support Agent Capstone

Design a tool-calling agent that can look up information and take constrained actions through explicit function calls, with guardrails.

Tracks: AI, LLMs, RAG & Agents

Open in Project Studio

Objectives

Define a small set of typed tools/functions an agent can call
Implement an agent loop that plans, calls tools, and observes results
Add guardrails: input limits, allowed actions, and human-readable audit logging
Reason about cost, latency, and failure modes in production

Before you start

Milestones

0%

1. Tool design

Define 2-3 tools with clear typed inputs/outputs.

Each tool has a name, description, and typed parameters
Tools only perform safe, constrained actions (e.g. read-only lookups)
Tool errors return structured failure information, not crashes

2. Agent loop

Implement plan → act → observe → respond.

The agent selects a tool based on the user's request
Tool results are incorporated into the final answer
The loop terminates within a bounded number of steps

3. Guardrails and observability

Add safety limits and basic observability.

Input length and request-rate limits are enforced
Every tool call is logged in an auditable way
The agent refuses actions outside its defined tool set

References

Anthropic: Tool use overview