Data science and machine learning applied to your operations

Suez guarantees its local authority clients a network efficiency figure calculated every day. Yet consumption readings arrive incomplete, between remote-reading failures and meters with no remote reading at all. Galadrim built the estimation engine that fills those gaps. It predicts consumption for meters with a history, taking into account the weather and neighbouring consumption. It rebuilds a typical profile for the least-equipped meter fleets. And it returns an overall efficiency figure with its margin of uncertainty. The engine runs on Suez's Google Cloud infrastructure.

Castrol, part of the BP group, runs an automated filling line on a fully offline site. Anyone entering the line mid-cycle risks a serious accident. We trained a detection model on several hours of video and 500 annotated images, chose hardware able to run it on site, then installed the system on the line. A camera watches the danger zone continuously: as soon as someone steps in, the alert goes off immediately, and no image ever leaves the plant.

Chanel's neuroscience unit wanted to understand how its products are really used, based on filmed tests. The goal was quantifiable data on how a person applies a cosmetic. Galadrim developed image recognition algorithms that track the user's head, hand and fingers throughout the video. Each gesture is counted and classified by type, such as a semi-circular massage or a light tap. That gives Chanel an objective measure of how a cosmetic is applied.

Every day, Vinci Construction assigns workers, plant machinery and locomotives to keep its rail operations running. The rules pile up: rest periods, whether a worker is certified to drive a given machine, the sequence of journeys from one site to the next. Galadrim built the planning engine inside its internal tool, which handles every worksite in a single calculation. A site manager opens the tool in the morning and finds the team schedule already built, compliant with every rule and ready for sign-off.

Groupe Blondel runs logistics for the Chantiers de l'Atlantique shipyard, where cranes, lifts and goods hoists set the pace for the whole site. The equipment had to run at full capacity while teams were assigned and package flows tracked, without queues building up in front of an overloaded machine. Galadrim built the application that brings all these constraints together. A model predicts upcoming package volumes to anticipate the real workload. Works supervisors book their lifting slots from the interface. And the assignment algorithm builds schedules that respect working hours and operator–machine compatibility. Every employee receives their own schedule in the morning.

Spoticar, the Stellantis group's used-car brand, runs a dealer network and the listings portal that brings dealers their buyers. The information that drives a purchase was spread across six systems that didn't talk to each other: stock, maintenance, manufacturer documentation, customer reviews and search histories. Galadrim brings them together in the group's single database, working within its ingestion framework and security rules. We then build the data dictionary that describes this data in the business's own vocabulary. A buyer asks a question in French and gets the vehicles that match; a dealer manages their stock; both work from the same data.