Malignant melanoma (MM) is the second most common cancer among active-duty U.S. military service members, who have higher incidence compared to civilians. Diagnosing MM presents significant challenges due to overlapping features with benign melanocytic lesions (BMLs), especially for rare subtypes and among the understudied adolescents and young adults. Notably, 92% of the military population are young adults (less than 40 years old). Melanoma misdiagnosis can lead to delayed treatment or overtreatment. While artificial intelligence (AI) holds great promise for diagnosing and prognosticating MM and BML, existing AI models are often trained on limited, homogeneous datasets from single institutions, restricting their accuracy and generalizability. Our project directly confronts these limitations by harnessing the extensive resources of the Department of War's Joint Pathology Center (JPC), the world's largest human tissue repository, which, since 1917, has accumulated biospecimens submitted for consultation from military, veteran, other government, and civilian medical centers, as well as biospecimens acquired through the Base Re-Alignment and Closure Act. This repository is enriched with rare, diagnostically complex cases from various military and civilian sources, making it the ideal foundation for developing highly robust and broadly applicable AI diagnostic and prognostic tools. However, JPC’s tissue repository data are unstructured and contain protected health information and difficult to query and summarize. Therefore, our project’s foundational aim is to employ AI-driven Natural Language Processing (NLP) techniques to de-identify, clean, structure, and harmonize this vast repository of data for cases of MM and BMLs in JPC’s Tissue Repository that were diagnosed in the past 30 years. This critical first step will transform millions of unstructured data points into a relational, queryable, and user-friendly database built on standardized terminologies and interoperable data models, enabling efficient retrieval and cross-cohort analyses. To help build this database, we have identified records for more than 18,000 MM cases and almost 38,000 BML cases diagnosed in the past three decades. Our second aim is to develop a Digital Dermatopathology Image Warehouse that contains whole-slide images of selected MM and BMLs linked to their corresponding diagnostic and dermatopathologic metadata. We aim to assemble a representative cohort of 1,500 MM and 1,000 BML cases, with an oversampling of rare and diagnostically challenging histological melanoma subtypes, such as Spitz, desmoplastic, nail unit, mucosal, and ocular variants. By the time of this poster presentation, we aim to present the initial results of our NLP de-identification algorithm and showcase our progress in making this unique dataset secure, accessible, and ready for research. Ultimately, our project will create two enduring national resources governed by the JPC: a comprehensive de-identified database of MM and BML cases in the JPC’s Tissue Repository, and a Digital Dermatopathology Image Warehouse, set to accelerate melanoma research for the entire scientific community for years to come. These resources will make it easier to query and summarize the cases of MM and BMLs in JPC’s Tissue Repository and will enable the development of image-based AI tools that can aid in diagnosis and prognosis especially of rare and diagnostically challenging MM and BML subtypes.