A VALUED CUSTOMER OF THIRDEYE DATA
Xperi: bringing hundreds of machine learning models under one platform
Xperi develops and licenses technologies used in entertainment and smart devices, with machine learning embedded across its products. Over years of work its teams had accumulated hundreds of models covering prediction, object detection and tag identification, built at different times by different people for different products.
LOCATION
San Jose, California, United States
COMPANY SIZE
Mid-scale Enterprise
INDUSTRY
Information Technology › Explore solutions for this industry
MLOps platform
Hundreds of models is an asset on paper and a liability in practice. Nobody could say with confidence which version of which model was running where, what data it had been trained on, or whether a model still performed the way it had when it was signed off. Retraining meant reconstructing the conditions from memory. New work often meant rebuilding something that already existed somewhere because finding it was harder than writing it again.
ThirdEye Data built the platform that put all of it under one system. That meant three connected pieces of work: designing and building the data lakehouse where training and operational data live, building the pipelines that bring data in and get it into usable shape, and establishing the model lifecycle itself, covering development, training, deployment and versioning.
The platform serves three different audiences with different needs. Data scientists need to experiment without breaking anything. Data engineers need the pipelines to be reliable and observable. Business stakeholders need to know what is running in production and how it is performing. Designing for all three rather than for whoever shouted loudest was the part that took judgement.
Hundreds of previously scattered models brought under a single managed lifecycle.
Related solution: MLOps and machine learning operations
Most organisations reach a point where the number of models they run exceeds the number they can actually account for. It usually arrives quietly. Models get built for specific problems, they work, they go into production, and the person who built each one is the only person who fully understands it. The failure is not dramatic. It shows up as an inability to answer simple questions about what is running.
ThirdEye Data builds the operational layer that makes those questions answerable: where data comes from and what state it is in, how a model gets from experiment to production, how versions are tracked, and how performance is monitored once something is live. We build it around how the organisation actually works rather than imposing a process, because an MLOps platform that data scientists route around is worse than none at all. Xperi is the largest engagement of this kind we have delivered.
DO YOU HAVE A SIMILAR USE CASE?
Talk to our team to learn more about what we did for Xperi, and how that experience applies to your own environment.
Explore What We Did for Other Customers
