Scientific Manuscripts

Design and Implementation of a Multi-Model AI-Based Content Moderation System in a Production Environment

An empirical study on the production-grade deployment of a fault-tolerant, multi-model AI moderation system powering the FindWay platform. The architecture decouples user interactions from heavy analytical tasks via an asynchronous distributed queue pipeline. It integrates an ensemble of 8 specialized NLP models (v3. v3.1–v6. v6.1), an EfficientNet-B0 computer vision module, and ArcFace biometric embeddings. Governed by a Policy Layer with Weighted Confidence Thresholding, the system prioritizes risk control and false-positive minimization—achieving 94.5% Precision in text and 91.8% in vision under continuous sub-second latency.