feature-store-validator

What problem does it solve?

This Skill eliminates the risk of silent feature store failures that cause production machine learning models to underperform or fail without warning in offline validation metrics. Uncaught issues like training-serving skew, point-in-time data leakage, and unmonitored feature drift can lead to incorrect predictions, financial loss, and damaged user trust before teams detect the problem.

Core Features & Use Cases

• Critical Failure Mode Auditing: Scans feature stores for 11 high-impact failure modes including training-serving skew, missing point-in-time joins for training data, online/offline value inconsistency, unmonitored feature drift, missing feature ownership, and unpinned feature versions at inference.
• Cross-Platform Validation: Supports auditing for Feast, Tecton, Hopsworks, Vertex AI Feature Store, Databricks Feature Store, and AWS SageMaker Feature Store, with platform-specific best practice checks for each.
• Use Case: For a team deploying a fraud detection model, this Skill would catch if the training pipeline uses a naive join that leaks future transaction data into features, or if the online feature store values don't match offline training values, preventing silent model degradation in production.

Quick Start

Invoke the feature-store-validator skill to audit your feature store for critical flaws that could cause production ML model failures.