If the catalogue defines what markets exist, the models turn them into live odds. Those models ran on individual machines, slow to deploy and impossible to govern. I owned the platform that brought them into one place.
The odds models, the things that actually price every market, were scattered across individual machines. One model per machine, no central hosting, no version control, no quality oversight. Deploying a new model took weeks of manual work, nobody could say with confidence which version was running where, and there was no governance layer to catch a bad model before it priced real markets. For a business whose product is the accuracy of its odds, that is operational risk sitting in plain sight.
This wasn't the catalogue problem again. The catalogue is about definitions, self-serve and static. Hosting is about compute and operations: models that run continuously under live odds load, that have to be versioned, governed, and consumable by downstream systems in real time. The work was designing an architecture that could host many sport models centrally and reliably, wiring it to the catalogue so markets flow into models automatically, and laying a governance layer over the top without slowing the teams who ship models.
I defined the product vision for centralised hosting and drove the migration off the siloed machines into one governed platform. Models became consumable through stateful APIs and real-time streams, versioning and quality oversight existed for the first time, and the hosting layer connected directly to the Markets Catalogue, so a market added in the catalogue maps into the relevant sport model automatically rather than through a manual handoff.
The technical migration was the easy half. The real work was governance teams would adopt rather than route around, which meant designing the consumption layer as carefully as the compute. Hosting and the catalogue together prove the modular platform holds: two products, owned separately, that compose into something neither could be alone.