We present a hybrid AI framework for quantifying short-term motor adaptation through unsupervised analysis and narrative summarization. At its core is the Acute Gait Adaptation Index (AGAI), a composite metric that integrates Autoencoder-based reconstruction error, Transformer forecasting deviations, and trajectory drift via Dynamic Time Warping (DTW). Applied to trial-wise gait data, AGAI enables the clustering of subjects into low, moderate, and high adaptation profiles. An ablation study confirms that no single feature provides equivalent separability, underscoring the value of multidimensional integration. Compared to classical variability metrics such as sample entropy, AGAI captures richer temporal and structural information, offering a more interpretable and individualized assessment of motor adaptation. To further enhance interpretability, we incorporate a domain-informed Large Language Model (LLM) that generates subject-level summaries aligned with AGAI trends. Results demonstrate that the framework supports adaptive, data-driven monitoring and offers a generalizable architecture for interpretable time-series modeling in neuroscience and human movement analysis.