Walgreens' supply chain was bleeding revenue to disruptions nobody saw coming. I embedded with their ops and engineering teams, designed the platform, shipped it across 200+ stores, and stayed until they ran it without me.
Late deliveries. Inventory shortages. Unplanned outages. Walgreens had data scattered across SAP, Oracle ERP, and IoT sensors — but no unified picture of where the supply chain was actually breaking. Managers were reacting to disruptions after revenue was already gone.
The brief, in plain terms: stop being surprised. Find the failure points before they hit the shelf.
I ran structured discovery workshops with Walgreens' ops and engineering leadership. The first question was always the same — where does the supply chain feel broken right now, and what does it cost when it breaks?
We mapped the actual process from purchase order to shelf, then identified the three highest-risk failure points. Everything that came next had to defend itself against those three.
SAP, Oracle, and IoT sensor data integrated into Microsoft Fabric OneLake as a single source of truth. Scikit-learn anomaly-detection models trained on 18 months of historical supply events — looking for the patterns that preceded disruptions, not the disruptions themselves.
Every model version was evaluated against historical ground truth before it ever showed up in front of a stakeholder. No model graduated to production until the numbers were real.
I don't ship and disappear. I built the runbook, documented every model decision, and ran training sessions with the customer team until they could operate the platform independently.
Anomaly detection components were packaged for reuse on the next engagement. The platform now runs without the delivery team.