Multi-source Data Ingestion for Warfighters -- 25-037

In todayís complex maritime battlespace, providing watchteams with actionable information derived from diverse data is crucial for improving operational effectiveness and outcomes. However, this can be challenging given the sheer volume and diversity of data, and the userís ability to manually locate, process, and identify key information in time constrained situations.ÝMetron will use Retrieval Augmented Generation (RAG) enabled solutions to build and demonstrate a tool that quickly surfaces critical information to the warfighter. The Smart Tactical AI Research System (STARS) is a human-centric, mission-focused application that sources textual and image documents, and internal and external databases, from available authoritative sources to answer operator queries.The Retrieval Augmented Generation, a form of Generative AI, utilizes the ability of Large Language Models (LLMs) to synthesize and summarize large amounts of unstructured data to answer questions or generate derived products and augments it by explicitly providing source documents in the generated responses. By explicitly providing source document text, as well as the userís query, to the LLM, RAG systems can include source citations in their responses, allowing users to verify information provided.ÝSTARS is comprised of two primary and independent subsystems ñ data ingest and chat, that are networked but can be operated separately and supports operations in denied, disrupted, intermittent, and/or limited (DDIL) environments.Authoritative data sources are critical to the utility of a RAG system. However, these foundational knowledge databases are not always current and do not include all relevant information. STARS will include a crawler that is configured to other source documents. Although canonical data is the ideal for systems like this, occasionally data sources will conflict on the same information. This may be because of incomplete primary information or more recent updates overriding previously published documents. STARS will provide intuitive responses to notify the user of conflicts.Development of STARS will include innovative designs required to meet the unique mission needs of military users, and the hardened system architectures that support them. By leveraging our deep understanding of Retrieval-Augmented Generator systems and the expected user queries and critical data sources that support them, we will deliver tailored solutions that improve warfighter effectiveness.