NB.
All work
05Commissioned build2026

OpsPilot

An AI operations inbox that turns messy business email into auditable, reviewable cases.

Recorded walkthrough · 1:22 · Silent recording
Role
AI engineer
Year
2026
Status
Demo runs on synthetic data
Stack
AgentsRAGHuman-in-the-loop

The mess

Support requests, leads, invoices, and documents arrive through the same inbox but require different handling. The task was to classify and route that intake using company policy while keeping uncertain or risky decisions reviewable.

The system

I built a commissioned operations inbox that turns email and documents into auditable cases. It classifies intent, extracts structured facts, retrieves versioned company knowledge, and proposes the next action with supporting evidence. Uncertain or risky cases enter a human review queue, and the case retains its decision history. The public recording uses synthetic data.

The result

A shared intake workflow that combines classification, policy evidence, proposed actions, and human review in one case history.

Policy evidence alongside proposed actions
Risky calls routed to human review
Full audit trail on every decision