3.3 AI Image Recognition and Model Risk

Parax integrates an ensemble of computer vision and machine learning models specialized in assessing asset authenticity across various categories, including luxury watches, fine art, sneakers, and rare collectibles. These models are trained on proprietary, expert-curated datasets to recognize brand-specific features such as signature markings, wear patterns, edge geometry, logo precision, and serial number placement.

To mitigate inherent model risk, Parax utilizes ensemble averaging, anomaly detection, and confidence thresholding. Assets scoring above predetermined confidence levels may bypass manual review in Tier III, while Tier II assets undergo AI pre-validation followed by expert human verification.

All AI model decisions, including input hashes, output confidence scores, and decision parameters, are logged within the metadata schema. This enables auditability, retrospective review, and dispute resolution.