This work introduces an innovative AI-driven pipeline for automated 3D CT assessment of post-COVID lung damage, addressing critical limitations of conventional methods. The framework combines a dual-encoder 3D U-Net (Dice = 0.983, 3.5% more accurate than traditional region-growing), an adaptive HU classifier (−900 to − 201 HU) that reduces misclassification by 22% versus fixed thresholds, and a self-attention-based RALE scorer that eliminates manual feature extraction. The multi-task architecture achieves 98.3% classification accuracy (vs. 95.1% in rule-based systems) with 40% lower inter-observer variability in clinical validation (500 scans, κ = 0.87 vs. radiologists). Key innovations include anatomically-aware 3D processing for volumetric biomarkers superior to 2D analysis, integrated HU/semantic feature learning that replaces manual ROIs, and full automation that reduces reporting time from hours to minutes while maintaining DICOM compatibility. By leveraging self-attention mechanisms and joint optimization, the system outperforms single-task models while providing clinically actionable insights beyond conventional severity scores, establishing a new standard for quantitative post-COVID lung analysis.