Inbox Auto-Sorter & Tagger
Build an internal tool that reads each email in your shared inbox, classifies it by topic, suggests the right tags and owner, and lets a teammate confirm before anything is applied - so the inbox sorts itself instead of being one giant pile.
A private, login-protected web app where you load a batch of inbox emails, the AI classifies each into a topic and suggests tags plus a target owner, a teammate reviews and confirms or edits, and only then are tags applied and routing recorded - with a clean tagged + routed CSV export.
Before you start
- A free Vercel account
- A free Supabase account
- A free Resend account (optional - only for notifications)
- An export of recent inbox emails as CSV/Google Sheet (or paste a batch)
- Your team's real tag/topic list and who owns each topic
The problem this kills
Every morning someone opens the shared inbox and faces one undifferentiated pile: billing questions tangled up with shipping complaints, a hot sales lead buried under spam, a technical fire three screens down. Triage by hand is slow, it's inconsistent (two people tag the same email two different ways), and the emails that need a fast human answer get found last. The inbox is the front door of your whole team, and right now it's chaos.
This tool turns that pile into a sorted, tagged, routed list - without handing the keys to a black box. The AI does the boring first pass; a person stays in control of what actually gets applied.
What you'll build
A small, private web app - just for your team, behind a login - that does this:
- You load a batch of inbox emails (from a CSV, a Google Sheet, or pasted in).
- The AI reads each one and classifies it into one of your topics (billing, technical, shipping, sales, spam - whatever your real list is), suggests tags from your tag list, and proposes who should own it.
- Low-confidence guesses land in a clear "needs human" bucket instead of being forced into a wrong box.
- A teammate reviews each suggestion, confirms or edits it, and approves.
- Only after approval are tags applied and the routing recorded - and you export a clean, tagged, routed email list in the exact columns your system expects.
What's inside the Implementation Plan
- It starts by interviewing you about your business. Before it builds anything, the plan has the AI agent ask about your current triage process, your actual topics and tags, how you name and route things, your typical and peak email volumes, and your messy edge cases. It reflects a short tailored spec back to you, you give it a thumbs-up, and then it builds a tool shaped to how you really work - not a generic template.
- A step-by-step build, where each step ends with a ready-to-paste prompt for your AI coding agent.
- A data model and validations driven by your answers - including locking the AI to your real tag list so it can't invent labels.
- The full governance layer (below) wired in from the start.
- A verification checklist so you know it actually works, plus a no-integration CSV fallback so you can ship today.
The governance it includes (this is the point)
This isn't a toy that auto-fires changes into your tools. The plan builds in, every time:
- Login so only your team can use it.
- Row-level security so people only ever see their own organization's data.
- A complete audit trail - who classified, who approved, what changed, and when.
- A hard human-in-the-loop gate - the AI only ever drafts tags and routing; a person reviews and approves before anything is committed.
- Duplicate guards keyed on each email's message-id, so the same email can't be processed twice.
- A validated tag list - the AI must pick from your approved labels and route low-confidence items to a "needs human" bucket.
Who it's for
Shared-inbox teams and leads who triage by hand every morning - support, ops, sales, or any group whose work lands in one common mailbox. If you can describe your topics and tags, you can build this. No coding background required.
You've got this - paste the first prompt and let the agent interview you.