As part of sharpening my AI automation skills, I built a workflow that handles customer support emails for an e-commerce shoe store scenario. The problem: support inboxes mix order updates, product questions, and refund requests together, and each type needs a different tone and different information to respond properly. Here is how the workflow works: (1) Gmail Trigger: Monitors the inbox and picks up new emails as they arrive. (2) Data Organization: Extracts and structures the key details from each email: sender, subject, and body. (3) AI Classification: A Text Classifier node reads the email and sorts it into one of three categories: Order, Inquiry, or Refund. (4) AI-Generated Replies: Powered by Gemini 3 Flash (free API from Google AI Studio for practice), the system applies a category-specific prompt so each reply matches the right tone and includes the right details, such as tracking info for orders or policy details for refunds. (5) Draft Creation: Instead of sending automatically, every AI-generated reply is saved as a Gmail draft. This keeps a human in the loop to review and approve responses before they reach a customer. My goal with this project was to practice designing multi-branch AI workflows that combine classification, context handling, and generation while keeping human oversight in the loop. This balance matters a lot when AI is applied to real customer-facing tasks. Built in n8n. Note: I'm using the free 14-day trial of n8n Cloud, and while building this workflow, the trial period had only 9 minutes left, so I couldn't record a video of it. But it runs smoothly and perfectly. I'll record a video of my next workflow.