PayPal's Cosmos.AI Platform is an enterprise-scale ML/AI platform designed to streamline end-to-end Machine Learning Development Lifecycle (MLDLC) across the organization.
•The platform reached GA in mid-2022 and now serves thousands of data scientists daily, with hundreds of deployed ML models handling tens of billions of inferencing requests per day
•Cosmos.AI adopts a 'best-of-breed with abstraction' design, decoupling capabilities from implementations (e.g., distributed training can run on Spark on-premises or as Vertex AI jobs on GCP)
•Multi-tenancy is built from the ground up to provide team autonomy, fine-grained access control, and data protection across the enterprise
•The unified platform covers the full MLOps lifecycle: data wrangling, feature engineering, model training, CI/CD, deployment, explainability, and drift detection
•Generative AI support including LLM fine-tuning, adapter hosting, and LLMOps was added as a horizontal extension on top of the existing platform