Federated Open-source Reasoner for Content Extraction (FORCE)

As the information landscape grows increasingly complex, the U.S. Marine Corps requires advanced capabilities to harness insights from vast, unstructured open-source data to feed the Intelligence Preparation for the Battlespace (IPB) process that is applied as part of new and ongoing operations. This work introduces a federated system that autonomously crawls distributed, unstructured data sources and transforms the information into structured formats optimized for downstream algorithmic processing. The architecture employs (locally hosted) large language models (LLMs) (E.g.,áhttps://www.llama.com/) and machine learning (ML) pipelines to extract entities, relationships, and context from natural language content, enabling real-time synthesis of actionable intelligence. Operating in a federated manner, the system respects data locality, reduces latency, and enhances operational security while maintaining scalability across heterogeneous data environments. The structured outputs are designed to feed directly into decision-support algorithms, enhancing mission planning, threat analysis, and strategic response. Our proposed approach, FORCE (Federated Open-source Reasoner for Content Extraction), offers the Marine Corps a resilient, AI-enhanced framework for turning fragmented open data into coherent, algorithm-ready, mission-relevant intelligence.There are key benefits of a federated system over enforcing a common format for all downstream algorithms, especially in the context of processing unstructured data for decision-making systems like those used by the Marines. A federated system will centrally store processed data aligned to a common data format, while allowing it to be shaped into the form each algorithm needs, thereby eliminating the need for downstream systems to adhere to the formats used in FORCE. Thus, legacy algorithms need not be modified to work with the system, andánew algorithms can be easily plugged in without requiring reworking of the data pipeline.