Assessing construction productivity objectively and in real-time remains challenging. This study proposes an automated framework leveraging Building Information Modeling (BIM) interaction logs and predictive modeling. The methodology involves systematic log data processing and feature engineering, generation of pseudo-label productivity scores derived from an initial expert-informed model, and a rigorous comparative evaluation of predictive architectures. Sixteen diverse models were evaluated using 10-fold cross-validation on a 500-instance dataset derived from construction logs. The cross-validation identified XGBoost as the top-performing architecture (R2 = 0.97 ± 0.01), demonstrating the effectiveness of gradient boosting on the engineered tabular features. The framework incorporates an integrated interface with visualization and natural language processing for enhanced insight generation and accessibility. While acknowledging limitations concerning pseudo-label usage and initial data processing steps, this research presents a robust, validated methodology for data-driven productivity assessment, offering a scalable alternative to traditional methods in construction project management.