Opalina Technologies
All work
Case 17
Manufacturing

AI for a Factory Whose Data Cannot Leave the Building

For a manufacturer with a strict zero-data-egress rule: a fully on-premise AI stack — demand forecasting, unified analytics, and sales-to-procurement automation — where not one byte travels to a cloud.

Private & On-Prem AI
ML Forecasting
A perfume manufacturer, Mexico
AI for a Factory Whose Data Cannot Leave the Building

The problem

Some companies' most guarded asset is not money but formulas. Our client, a perfume manufacturer running SAP Business One alongside a French formulation ERP, had a rule that disqualified nearly every AI vendor who walked in the door: no data leaves the building. Ever. Not to a cloud, not to an API, not "anonymised." Which left them watching competitors adopt AI while their own systems — two ERPs that barely spoke — kept planning blind.

The solution

Our design takes the constraint as the brief. Everything runs on hardware inside the client's walls: demand-forecasting models trained on their own sales history; an analytics layer that finally unifies the two ERPs into one picture of the business; automation that turns sales projections into procurement suggestions; and customer and supplier intelligence built from data that never moves.

Where language models are useful, they are local models on local machines — the same pattern as VerifAi's self-hosted mode, already working software. The solution demonstrates something we believe more clients will demand each year: modern AI does not require surrendering your data to someone else's computer. The intelligence can come to the data.

Built for enterprise

zero data egress — fully on-premise AI
locally hosted LLMs on client hardware (CPU/GPU)
dual-ERP unification (SAP Business One + formulation ERP)
demand forecasting on first-party data
sales-to-procurement automation
trade-secret-safe architecture
They wouldn't let their secret recipes near the internet — sensible. So we designed the brain to live inside their building, learning only from their own books.
In plain language

This story is relevant to

Manufacturers with trade secrets, defence suppliers, healthcare providers, and any enterprise whose compliance office has vetoed cloud AI.

Facing something like this?

The architecture, phasing, and commercials are already prepared. Most engagements begin with a proof-of-concept measured in weeks.

Book a call