Model Governance
As enterprises usher AI out of the lab and into live streaming applications, algorithms and data pipelines are entrusted with complex decision making. With the right tooling and care, AI can drive value to mission-critical business processes in ways that are fair, transparent, and secure. C3 AI Application Platform provides robust Model Governance capabilities along the end-to-end AI application development lifecycle. Built-in tooling for machine learning and data governance encourages best-practices with features like automated model scoring and interpretability to ensure quality checks are met, ML Pipelines and logging for transparency and reproducibility, and role-based access controls on all data and modeling assets for security.
As business and regulatory requirements evolve, administrators need tooling that can adapt and grow to their changing needs. C3 AI Application Platform’s model driven architecture means each step, from data ingest, to model training, testing, approvals, model promotion, and retirement, is fully configurable and extendible.
Transparency
C3 AI Application Platform stores a complete time-based lineage of data and machine learn pipelines, which means the origins and internals of a model or prediction can be audited or replayed from any point in time.
AI Explainability
All C3 AI ML Pipelines are fully integrated with leading interpretability methods like Lime, SHAP, ELI5. Interpretability results are stored, making it easy to integrate Explainability directly to the live application.
Human-AI Collaboration
Assess and configure the right level of automation with “human-in-the-loop” AI supervision and decision making. Configure when to automatically retrain and promote models based on key performance metrics and checks, or set up an expert approval process.
Security
Meet rigorous security and governance requirements with rich metadata including your full model history, including model signing and verification, data lineage and versioning, and access control for all modeling artifacts (including notebooks, input data, features, models, and outputs).
Logging and Alerts
All user interactions with models and other APIs are centrally logged and monitored. Alerts are configured to signal data drift, performance issues, or anomalous events.