Built an AI-powered invoice and receipt processing system using Telegram, OCR, OpenAI, and Supabase. The workflow automatically extracts invoice data from images/PDFs, stores it in a database, and enables natural-language financial queries via chat.
An AI workflow that reads every incoming email, sorts it into 6 categories, and takes the right action automatically.
The problem: businesses lose time and leads because customer emails are mixed with promotions, job alerts, and security emails.
What I built:
Gmail Trigger that picks up each new email
AI Text Classifier (Google Gemini) with 6 categories: Security Alert, Promotions, Jobs, Personal, Payments, and Client Inquiries
Automatic Gmail labels for every category
Google Sheets log for jobs, leads, and payments
AI information extraction for client emails (name, phone, service, budget)
AI-written reply drafts for client emails, saved as drafts so a person approves every reply
Instant Discord alerts for security, payment, and client emails
Tools: n8n, Google Gemini, Gmail, Google Sheets, Discord
This project is a working demo built on my own inbox. The same system can be adapted to any business email, such as travel agencies, event venues, stores, recruiters, and consultants.
Congrats on getting it approved. Drafting instead of auto-sending is the right default. One edge case worth a guard: calls where the transcript comes back empty or cut short, otherwise the model writes a confident follow-up from almost nothing.