Skip to main content

Lab 18: Building a Smart Travel Planner Challenge

Goal​

Goal​

In this lab, you will build a Smart Travel Planner that uses the fan-out/gather pattern. You will use a ParallelAgent to concurrently search for flights, hotels, and activities, and then use a final agent to synthesize the results into a complete travel itinerary.

The Fan-Out/Gather Pattern​

This lab implements a powerful and efficient architectural pattern.

        ┌──── Agent 1 (flights) ────┐
User ───┼──── Agent 2 (hotels) ─────┼──→ Merger Agent → Final Result
        └──── Agent 3 (activities) ─┘

      ParallelAgent (fast!)       SequentialAgent (combine)

Step 1: Create the Project Structure​

  1. Create a new project:

    adk create travel-planner

    When prompted, choose the Programmatic (Python script) option.

  2. Navigate into the new directory:

    cd travel-planner

Step 2: Assemble the Fan-Out/Gather Pipeline​

Exercise: Open agent.py. The specialist agents (flight_finder, hotel_finder, activity_finder, and itinerary_builder) have been provided for you. Your task is to assemble them into a functioning fan-out/gather pipeline.

# In agent.py (Starter Code)

from __future__ import annotations
from google.adk.agents import Agent, ParallelAgent, SequentialAgent

# ===== Specialist Agents (Provided for you) =====

flight_finder = Agent(name="flight_finder", model="gemini-2.5-flash", ..., output_key="flight_options")
hotel_finder = Agent(name="hotel_finder", model="gemini-2.5-flash", ..., output_key="hotel_options")
activity_finder = Agent(name="activity_finder", model="gemini-2.5-flash", ..., output_key="activity_options")
itinerary_builder = Agent(name="itinerary_builder", model="gemini-2.5-flash", ..., instruction="...{flight_options}...{hotel_options}...{activity_options}...")

# ============================================================================
# FAN-OUT: PARALLEL DATA GATHERING
# ============================================================================

# TODO: 1. Create a `ParallelAgent` named `parallel_search`.
# TODO: 2. Add the three "finder" agents to its `sub_agents` list so they
# run concurrently.
parallel_search = None

# ============================================================================
# COMPLETE FAN-OUT/GATHER PIPELINE
# ============================================================================

# TODO: 3. Create a `SequentialAgent` named `travel_planning_system`.
# This will be the main entry point for the entire workflow.
# TODO: 4. Add the `parallel_search` agent as the first step and the
# `itinerary_builder` agent as the second step.
travel_planning_system = None

# TODO: 5. Set the `root_agent` to be your `travel_planning_system`.
root_agent = None

(Note: The full agent definitions are in the lab-solution.md if you need to inspect them, but you don't need to change them for this exercise.)

Step 3: Run and Test the Pipeline​

  1. Set up your .env file.
  2. Navigate to the parent directory (cd ..) and start the Dev UI: 
    adk web travel-planner
  3. Interact with the pipeline:
    • Send a travel request, like: "Plan a 7-day vacation to Honolulu".
  4. Examine the Trace View:
    • Expand the trace to see the SequentialAgent run.
    • Inside, expand the ParallelAgent to see the three finder agents running concurrently.
    • Observe the itinerary_builder running after the parallel step is complete.
    • Note: Notice in the trace that the three finder agents start at roughly the same time. The total time for this step is determined by the slowest of the three, not the sum of all three, which is much more efficient!

Having Trouble?​

If you get stuck, you can find the complete, working code in the lab-solution.md file.

Lab Summary​

You have successfully built a high-performance, multi-agent system using the fan-out/gather pattern. You have learned:

  • How to configure and use a ParallelAgent to run agents concurrently.
  • How to combine ParallelAgent and SequentialAgent to create efficient data-gathering and synthesis pipelines.
  • How to verify parallel execution in the Trace View.

Self-Reflection Questions​

  • What would be the performance impact if you replaced the ParallelAgent in this lab with a SequentialAgent?
  • It is critical that the three finder agents have different output_keys. What would happen if they all had the same output_key, like "search_result"?
  • Can you think of another real-world problem (besides travel planning) where the fan-out/gather pattern would be a highly effective architecture?

🕵️ Hidden Solution 🕵️​

Looking for the solution? Here's a hint (Base64 decode me): L2RvYy1hZGstdHJhaW5pbmcvbW9kdWxlMTgtcGFyYWxsZWwtd29ya2Zsb3ctYWdlbnRzL2xhYi1zb2x1dGlvbg==

The direct link is: Lab Solution