Building Multi-Agent Systems

🚀 What I Built
The system solves a common problem: turning an unclear intention (“I need to email a client about a delay”) into a clear, well‑written email. Instead of relying on a single prompt to do everything, the system breaks the task into three focused agents. This reflects the track’s emphasis on specialization and distributed orchestration, where each agent contributes one piece of the solution .

The user enters a goal and selects a tone, and the system walks through the workflow step by step, showing how each agent transforms the message.

🛰️ Live Cloud Run App
Paste your deployed Cloud Run embed here:

Code
<iframe
src="YOUR_CLOUD_RUN_URL"
height="600"
width="100%"
style="border:1px solid #ccc; border-radius:8px;"

This satisfies the requirement to embed the working web app directly into the submission, as outlined in the track’s Part 3 instructions .

🤖 How the Agents Collaborate
The system uses three independent Cloud Run microservices, each representing a specialized agent. This mirrors the examples provided in the track, such as the Email Drafter pattern where Topic → Writer → Editor form a natural pipeline .

Topic Agent — Interprets the user’s goal and proposes a subject line and outline.

Writer Agent — Expands the outline into a full draft shaped by the selected tone.

Editor Agent — Refines clarity, tone, and flow to produce the final polished email.

Each agent receives structured JSON, performs its role, and passes the result to the next stage. The frontend orchestrates the sequence and displays each step so the user can see the workflow unfold.

🗺️ Architecture Diagram
Code
User Input (Goal + Tone)
|
v
[ Topic Agent ] — Cloud Run microservice
|
v
[ Writer Agent ] — Cloud Run microservice
|
v
[ Editor Agent ] — Cloud Run microservice
|
v
Final Polished Email (shown in UI)
This reflects the distributed architecture described in the track’s learning objectives, where each agent has a focused responsibility and communicates through clear interfaces .

🖼️ Screenshots
Add your screenshots here:

Input panel (goal + tone)

Stepper showing Topic → Writer → Editor progress

Final email output

Any additional UI elements you want to highlight

📚 Key Learnings
Working through the track and building the system reinforced several important concepts emphasized in the curriculum:

Specialized agents are more reliable than monolithic prompts. Breaking the task into roles made the system easier to debug and reason about.

Distributed systems require clear contracts. Passing structured JSON between agents highlighted the importance of consistent interfaces.

Cloud Run makes modular deployment straightforward. Each agent runs independently, scales automatically, and stays isolated.

UI transparency builds trust. Showing each agent’s output helps users understand how the system works and why the final email looks the way it does.

The multi‑agent mindset is powerful. Thinking in terms of roles and responsibilities opens up new ways to design AI‑driven applications.

These reflections align with the track’s Part 3 goal of documenting your architecture and sharing what you learned with the community .
https://codepen.io/editor/Dancodepen-io/pen/019cf78b-42e9-7afb-9599-0a3cd9300621