<p>Modern health analytics increasingly relies on large volumes of high-dimensional training logs, collected from various wearable devices, sensor-equipped gym equipment, and mobile fitness applications. While these data streams offer unprecedented opportunities for competitive training performance optimization, injury prevention, and personalized training guidance, they also present significant computational challenges. Existing approaches often struggle with the soaring complexity and real-time requirements of such data, making brute-force methods infeasible and more sophisticated indexing structures resource-intensive. In this paper, we address these challenges by proposing an edge computing-enabled scalable Locality-Sensitive Hashing (LSH) approach for large-scale similarity search in training log embeddings. By designing a hash-based index that captures core data properties of physical training logs stored in distributed edge servers, our method significantly reduces the computational overhead associated with nearest neighbor searches across millions of workout records. We illustrate how this pipeline integrates into real-world health analytics systems to facilitate rapid retrieval of comparable sessions, enabling personalized workout recommendations and detailed performance insights. Experimental evaluations on synthetic datasets show that our LSH-based framework achieves high retrieval accuracy while maintaining low computational time and memory consumption, thus offering a scalable solution to the challenges of modern health data.</p>