AI Email Response Generator
Built for Hollow Technologies to handle a support inbox receiving 60 to 80 emails a day across three people who were already behind. The workflow catches every new email, sends it to Gemini for classification, logs the ticket to a sheet, then drafts a reply that matches the company's tone and saves it directly to Gmail. Nobody writes from scratch anymore.
The problem
Hollow Technologies was averaging 18-hour response times on support emails. Not because the team was slow but because 60 to 80 emails a day across three people is just math. Feature questions, bug reports, onboarding help, billing queries, all landing in the same inbox, all waiting for someone to read, process, and write a reply from scratch.
The team knew what good responses looked like. They just didn't have time to write them.
What this workflow does
Every new email triggers the workflow automatically.
Gemini reads the email and classifies it: category, priority, and a one-line summary of what the sender actually needs. That gets logged to Google Sheets so the team has a full ticket record even if something goes wrong downstream.
Then Gemini drafts a reply. Not a template. A context-aware response written in Hollow Technologies' tone, warm and clear without being robotic. The draft lands directly in Gmail, threaded to the original email, ready to review and send with one click.
Every email classified, prioritised, and summarised before anyone reads it.How it's built
Seven nodes.
Gmail triggers on every new support email. A wait node spaces out the Gemini calls to stay within rate limits. Gemini classifies the email and returns structured JSON with category, priority, and summary. A code node parses that response. Google Sheets logs the ticket. A second wait node runs before the next Gemini call. Gemini drafts the reply. Another code node pulls the clean draft text. Gmail saves it as a draft threaded to the original.
The two-step Gemini approach matters. Classification and drafting are separate calls with separate prompts. Combining them into one call produces worse output on both fronts.
See it running
The result
The support team stopped writing emails from scratch. Every incoming message gets an immediate classification and a ready-to-send draft. Response times dropped not because the team got faster but because the slow part was removed.
Is your support inbox running your team instead of the other way around?
Tell me the inbox volume, the categories you deal with, and the tone you want. I'll build something that handles the drafting so your team only has to approve and send.
More projects
AI Social Media Repurposer
Paste a blog URL, get a LinkedIn post, a tweet, and an email snippet back. All three land in a Google Sheet with the source URL and timestamp.
AI CV Screener
Reads every CV in a Google Drive folder, scores each one against the job criteria, and outputs a ranked list to Google Sheets. No manual shortlisting.