Help build the dataset behind a student email assistant — ~30 minutes.
1What this is (please read)
We're building an AI assistant that triages student email. To measure whether it
actually works, we need real, labeled emails from real students — that's you.
What's collected: a random sample of emails (subject, sender, body text) from an inbox you connect, across the school year.
Access is read-only. Nothing is sent, moved, marked read, or deleted in your inbox.
You choose what's included. You'll review every email before finishing — remove anything personal with one click, no questions asked.
Where it goes: a private dataset used only to evaluate triage models. Never published, never shared beyond the project.
Curious what the labels mean? Skim the labeling guide — it's two minutes and makes your review much faster.
2Connect an inbox
Connect your personal Gmail, your school Outlook, or both. Access is read-only and lives only in this session.
School (.edu) account blocked from signing in? Paste a token instead
Open Graph Explorer and sign in with your school account.
Run the default GET /me query once; if prompted, consent to the Mail.Read permission (click your avatar → Consent to permissions if you need to find it).
Open the Access token tab under the query box and copy the whole token.
Paste it below. It's valid for about an hour — plenty for one session.
3Sample your emails
We pull a random sample spread across the school year (not just recent mail),
then an AI takes a first pass at labeling each one so your review is quick.
4Review each email
Check the AI's suggested labels, fix anything wrong, and confirm. Remove any email you'd rather not share.
Not sure where an email belongs? See the labeling guide.
Nothing to review yet — sample your inbox above.
✓All done — thank you!
Every email is reviewed and your contribution is saved. You can close this tab.
Want to add your other inbox too? Just connect it in step 2 and sample again.