Autonomous AI Email Triage & Operations

AI EMAIL TRIAGE & AUTONOMOUS OPERATIONS

A hybrid workflow that understands incoming email, calculates urgency, enforces identity-sensitive policies, and executes bounded operational actions without an approval queue.

MOCK DATA NOTICE All email scenarios, senders, companies, and benchmark labels are synthetic demonstration data.

PROOF OF WORK • 500 deterministic email evaluations • 100% category and priority accuracy on the disclosed synthetic benchmark • 100% automatically routed or clarified • Deterministic replay passed • 7 automated tests passing

AUTONOMOUS AI PROOF The live 3B model classified 145/200 stress cases correctly by itself. Deterministic evidence checks and policy enforcement detected unsupported decisions and applied 55 automatic fallbacks. The complete system produced 200/200 correct final actions, including 100/100 multi-signal and prompt-injection stress cases, with 0 human approvals.

The final 100% result belongs to the complete bounded system, not the model alone. Identity-sensitive cancellation and account operations request verification automatically instead of executing unsafe actions.

HOW IT WORKS Email → normalization and weighted rules → category, urgency, and confidence → bounded action. Ambiguous language can use AI, but policy code independently validates evidence and selects the permitted action.

REPRODUCE THE EVIDENCE The repository includes labeled mock inputs, all 200 case decisions, machine-readable benchmarks, seven tests, and CI checks.

Repository: https://github.com/Milo318/ai-email-triage-poc