OZDEP - Data Engineering Platform

OZDEP - Data Engineering Platform

OZDEP covers the part of machine learning that happens before the model, and the part that happens after it. Data arrives however you have it: files you upload, connectors to external sources, live streams from APIs and devices, or datasets you generate synthetically. It lands in one governed store, where it is checked before anyone builds on it. OZDEP looks for missing values, outliers, and duplicates, and also for personal or health information sitting where it should not be, and images too dark or too compressed to train on. Datasets are cleaned and reshaped in place, and each one carries a visible state as it moves from newly registered to validated to trusted. Nothing is lost along the way. Every dataset is versioned like code, with a history you can browse, compare, and roll back to. Every model traces back to the data and the run that produced it, and ownership, access, audit, and privacy requests are handled for each workspace. Training uses your own Python code, running on GPU clusters with data mounted for you and every metric and artifact recorded. Finished models go live as endpoints you can call, whether that is a large language model, an image generator, or a standard ML model. The infrastructure underneath comes from OZVIP, which provides the OpenStack virtual machines, Kubernetes and Ray clusters, and vLLM serving that let training and deployment scale. What OZDEP produces then feeds OZFORGE, which reads workspace datasets as live streams and calls deployed models to build digital twins.