Exercise 7 — Capstone Challenge: Build Your Own Agent System

Overview

You’ve learned the core concepts — agents, tools, structured output, multi-agent workflows, and DevUI. Now it’s time to build something on your own.

Choose one of the three challenges below and create a working solution from scratch. There is no starter code; use what you’ve learned in Exercises 1–6 as your foundation.

Challenge Options

Option A: Custom Function Tool + Agent

Build a Python function that an agent can call as a tool.

Example ideas:

• Weather lookup (mock data)
• Database query helper (SQLite)
• File analyzer (word count, structure, etc.)
• Calculator with history

Requirements:

1. Define at least one Python function with type hints and a docstring
2. Register it as a tool for an agent
3. Create an agent that uses the tool based on user input
4. Handle errors gracefully (invalid input, missing data, etc.)

Evaluation criteria:

• Tool is callable by the agent
• Agent invokes the tool correctly based on the conversation
• Output is meaningful and well-formatted

Option B: New Domain Multi-Agent Workflow

Design a multi-agent workflow for a completely different domain (not event planning).

Example ideas:

• Code review pipeline (lint → review → summary)
• Research paper summarizer (search → extract → synthesize)
• Travel itinerary planner (destination → activities → budget)
• Product launch checklist (market research → copy → timeline)

Requirements:

1. At least 3 specialist agents with distinct roles
2. Use WorkflowBuilder with edges connecting them
3. Include at least one tool (hosted tool or MCP)
4. Consume streaming events from the workflow execution

Evaluation criteria:

• Workflow executes end-to-end without errors
• Each agent produces relevant, domain-specific output
• Events are tracked and displayed during execution

Option C: Extend the Event Planning System

Add a new capability to the existing event planning workflow from Exercise 5/6.

Example ideas:

• Add a Transportation agent that plans logistics between venues
• Add structured output (Pydantic model) to the booking agent
• Create a DevUI entity that loads the extended workflow
• Add conditional routing (e.g., skip catering for virtual events)

Requirements:

1. At least one new agent or a significant enhancement to an existing one
2. Integration with the existing workflow pattern (see src/demo5_workflow_edges.py)
3. A DevUI entity that can be loaded (see entities/event_planning_workflow/)

Evaluation criteria:

• Extended workflow runs without breaking existing agents
• New capability produces useful output
• Entity loads correctly in DevUI (if applicable)

General Requirements (All Options)

Regardless of which challenge you choose, your solution must:

• Follow the repository’s established patterns:
  • Fill-only .env loading with dotenv_values (don’t override existing vars)
  • _require_env() for mandatory environment variables
  • async with for credentials and clients
• Handle errors with actionable messages (tell the user what to check/fix)
• Pass syntax checking: python3 -m compileall -q .
• Include a brief comment at the top of your main file explaining what the system does

Tips

1. Start by sketching your design on paper or in a comment block — list agents, tools, and data flow before writing code
2. Reuse patterns from Exercises 1–6 — copy boilerplate (.env loading, credential setup, _require_env) from the starter files
3. Test incrementally — get one agent working before adding more
4. Refer to existing solutions — src/demo1_run_agent.py through src/demo6_devui.py and entities/ are your best reference

Boilerplate Skeleton

Every solution should start with something like this:

"""
Capstone: <brief description of your system>
"""

import asyncio
import os
from pathlib import Path

from dotenv import dotenv_values

# ── .env loading (fill-only) ────────────────────────────────────
_DOTENV_PATH = Path(__file__).resolve().parents[1] / ".env"
_dotenv = dotenv_values(_DOTENV_PATH)
for _k, _v in _dotenv.items():
    if _v is None:
        continue
    _existing = os.getenv(_k)
    if _existing is None or not _existing.strip():
        os.environ[_k] = _v


def _require_env(name: str) -> str:
    value = (os.getenv(name) or "").strip()
    if not value:
        raise RuntimeError(
            f"Required environment variable is missing or empty: {name}. "
            "Set it via .env / export / Codespaces secrets and try again."
        )
    return value


async def main() -> None:
    # Your solution here
    ...


if __name__ == "__main__":
    asyncio.run(main())

Sharing

If time permits, present your solution to the group! Walk through:

• What challenge you chose and why
• Your design (agents, tools, flow)
• A live run (or key output)
• What you’d improve with more time

Solution Reference

There is no single correct solution for this exercise — your creativity is the solution.

For patterns and inspiration, review:

• src/demo1_run_agent.py – src/demo6_devui.py (working reference solutions)
• entities/event_planning_workflow/ (DevUI entity example)
• entities/ai_genius_workflow/ (alternative workflow example)