Laser grooving is a critical preparatory step in wafer dicing for advanced semiconductor packaging, ensuring reduced mechanical stress and improved dicing precision. However, groove quality assessment remains challenging due to profile irregularities and noise in depth measurement data. This paper presents an automated groove inspection system that leverages long short-term memory (LSTM) networks to analyze depth profiling data acquired via white light interferometry. The proposed method predicts the original wafer surface level and extracts key groove metrics, including groove depth, width, and debris zone characteristics. Unlike traditional rule-based or image-based methods, the LSTM-based model processes 1D cross-sectional height profiles directly, achieving a prediction accuracy of 99.87% under practical industrial tolerances. The system integrates seamlessly into semiconductor manufacturing pipelines, supporting real-time process monitoring, adaptive parameter tuning, and predictive maintenance. Experimental results validate the superiority of the proposed approach over conventional techniques, offering enhanced robustness, accuracy, and scalability for high-throughput wafer-level metrology.