a
Multi-Agent Research System — Workshop | AllCode
Scenario 03 · Hands-On Workshop

Build a Multi-Agent
Research System

In this workshop you will set up the Scenario 03 research pipeline, run it end-to-end, inspect each agent's output, extend the system with a new capability, and reflect on the design tradeoffs. Complete each step to unlock the next.

Steps7
Total Points100
Est. Time60–90 min
PrereqsPython 3.11+ · Anthropic API key
1
Step 1
2
Step 2
3
Step 3
4
Step 4
5
Step 5
6
Step 6
7
Step 7
1
Step 1: Set Up the Project
Clone the files and install dependencies
5 pts
setuppython

Download the Scenario 03 files from the teaching aid. You should have the following structure. If you copied the files manually, double-check the layout matches exactly — import paths depend on it.

scenario03/ ├── main.py # entry point ├── schemas.py # shared dataclasses ├── requirements.txt └── agents/ ├── __init__.py # empty — makes agents a package ├── search_agent.py ├── analyst_agent.py ├── synthesis_agent.py └── report_agent.py
⚠ Rename requiredIf you saved the report agent as eport_agent.py (missing the r), rename it now: mv agents/eport_agent.py agents/report_agent.py

Install the single dependency:

pip install anthropic

Set your API key. Pick whichever method matches your environment:

# Option A — shell export (current session only) export ANTHROPIC_API_KEY="sk-ant-..." # Option B — .env file (requires python-dotenv) echo 'ANTHROPIC_API_KEY=sk-ant-...' > .env pip install python-dotenv
✓ TipIf you use Option B, add from dotenv import load_dotenv; load_dotenv() at the top of main.py before the Anthropic import.
Completion Checklist
Files are in the correct directory structure shown above
agents/__init__.py exists (even if empty)
report_agent.py is correctly named (not eport_agent.py)
anthropic package is installed: python -c "import anthropic; print('ok')"
ANTHROPIC_API_KEY is set in your environment
0 / 5 tasks checked
2
Step 2: Verify the Imports
Complete the previous step to unlock
🔒
3
Step 3: Run Your First Research Pipeline
Complete the previous step to unlock
🔒
4
Step 4: Inspect the Agent Outputs
Complete the previous step to unlock
🔒
5
Step 5: Tune for a Failing Case
Complete the previous step to unlock
🔒
6
Step 6: Extend: Add a Gap-Fill Step
Complete the previous step to unlock
🔒
7
Step 7: Reflect on the Architecture
Complete the previous step to unlock
🔒