SupplyBrain

Option comparison
Recovery overview
Risk alert

Inspiration

A supplier may give a new delivery date by email while the purchase order still shows the old one. Checking that change against inventory, production, and customer orders takes time. We built SupplyBrain to show the risk while there is still time to respond.

What it does

SupplyBrain links supplier updates to the orders they may affect. In our fictional NorthStar Robotics case, a shipment of 400 drives moves from October 20 to November 10. The app finds a gap of 380 drives across the production plan, with 60 contracted robots at risk.

It compares recovery plans, shows their cost and delivery impact, and points out missing evidence. A manager can question the first recommendation and explore a plan that uses pilot stock and a smaller supplier purchase. Approval prepares drafts for review; it does not place an order.

How we built it

A React and Vite interface sits on a FastAPI backend. PostgreSQL stores business evidence and case history. LangGraph coordinates two Claude agents, while the Sumo Engine checks their results against calculated quantities, dates, costs, and constraints. Triggerware integration can report changes in watched tables. The app also starts a new analysis when evidence changes.

Challenges we ran into

The case unfolds in stages, so later supplier confirmations must not appear in an earlier decision. We also had to keep plausible AI explanations from changing the numbers. Evidence snapshots, scenario versions, and engine checks address both problems.

Accomplishments that we're proud of

The simulated hybrid plan protects all 120 contracted robots with $24,000 in upfront spending. The app still shows the 130 forecast robots left uncovered and the original Pacific purchase order that remains open.