<p>The domain of Reverse Engineering (RE) has undergone a radical transformation, evolving from a tool for geometric replication into an intelligent, data-driven process for design intent recovery and digital model generation. This evolution represents a core advancement in bridging engineering practice with computational innovation. This comprehensive review synthesizes decades of research to chart this progression, with a specific focus on the disruptive rise of Deep Learning (DL), which has initiated a paradigm shift from traditional geometric-fitting algorithms to semantic-aware, AI-driven reconstruction. We provide a structured, comparative analysis of techniques—from classical B-spline fitting and Delaunay triangulation to modern deep implicit functions and hybrid analytic-neural pipelines—evaluating their performance across diverse object types, data quality conditions, and practical industry requirements. Key trends identified include the inexorable move towards full automation, enhanced robustness to imperfect data, and the direct recovery of high-level, editable CAD structures, directly impacting design and manufacturing workflows. The article delineates current research gaps, such as the generalization of DL models and the need for explainable AI in engineering contexts and posits that the synergistic integration of geometric priorities with learned models will define the next era of intelligent reverse engineering, fundamentally bridging AI innovation with engineering practice.</p>