Skip to main content

Lab 17: Building a Blog Post Generator Pipeline Challenge

Goal​

Goal​

In this lab, you will build a multi-step content creation pipeline using a SequentialAgent. This will demonstrate how to chain multiple specialist agents together, passing data from one to the next.

The Pipeline Stages​

  1. Research Agent: Gathers key facts about a topic.
  2. Writer Agent: Creates a draft blog post from the research.
  3. Editor Agent: Reviews the draft and suggests improvements.
  4. Formatter Agent: Applies the edits and formats the final post in Markdown.

Step 1: Create the Project Structure​

  1. Create a new project:

    adk create blog-pipeline

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

  2. Navigate into the new directory:

    cd blog-pipeline

Step 2: Assemble the Pipeline​

Exercise: Open agent.py. The four specialist agents (research_agent, writer_agent, etc.) have been provided for you. Your task is to assemble them into a functioning pipeline.

# In agent.py (Starter Code)

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

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

research_agent = Agent(
    name="researcher", model="gemini-2.5-flash",
    instruction="...", # Gathers facts
    output_key="research_findings"
)
writer_agent = Agent(
    name="writer", model="gemini-2.5-flash",
    instruction="...writes a draft based on {research_findings}...",
    output_key="draft_post"
)
editor_agent = Agent(
    name="editor", model="gemini-2.5-flash",
    instruction="...reviews the {draft_post}...",
    output_key="editorial_feedback"
)
formatter_agent = Agent(
    name="formatter", model="gemini-2.5-flash",
    instruction="...applies {editorial_feedback} to the {draft_post}...",
    output_key="final_post"
)

# ===== Create the Sequential Pipeline =====

# TODO: 1. Create a `SequentialAgent` named `blog_creation_pipeline`.
# TODO: 2. In the `sub_agents` list, add the four specialist agents
# in the correct logical order: research -> write -> edit -> format.
blog_creation_pipeline = None

# TODO: 3. The ADK requires a `root_agent` to be defined.
# Set the `root_agent` to be your `blog_creation_pipeline`.
root_agent = None

(Note: The full agent instructions 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 blog-pipeline
  3. Interact with the pipeline:
    • Send a topic to write about, like: "the history of the internet".
  4. Examine the Trace and State Tabs:
    • Trace View: Expand the trace to see the SequentialAgent running its four sub-agents in order.
    • State View: After the run, inspect the state to see the output of each step (research_findings, draft_post, etc.).

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 deterministic, multi-agent pipeline. You have learned to:

  • Configure a SequentialAgent to orchestrate multiple sub-agents.
  • Understand how output_key and state variables ({key}) are used to pass data between agents in a sequence.
  • Analyze the execution of a pipeline using the Trace and State views.

Self-Reflection Questions​

  • The SequentialAgent is deterministic. What does this mean, and why is it a desirable property for a workflow like content creation?
  • What do you think would happen if you forgot to add the output_key to the research_agent? How would the writer_agent behave?
  • How could you modify this pipeline to include a human-in-the-loop? For example, what if you wanted a human to approve the draft_post before the editor_agent runs?

🕵️ Hidden Solution 🕵️​

Looking for the solution? Here's a hint (Base64 decode me): L2RvYy1hZGstdHJhaW5pbmcvbW9kdWxlMTctc2VxdWVudGlhbC13b3JrZmxvdy1hZ2VudHMvbGFiLXNvbHV0aW9u

The direct link is: Lab Solution