Omniscient
AFWERX · AFWERX STRATFI · AFWERX
Award
Description
Military supply chains are increasingly complex, requiring rapid, data-driven decisions to maintain operational readiness and sustain global logistics networks. Yet, the heterogeneity and siloed nature of available data, ranging from legacy maintenance records to real-time sensor feeds, create significant integration challenges and make it difficult for the Air Force to unify disparate data sources. Information about supply chains often resides in separate databases and formats, resulting in fragmented visibility across aircraft parts, vendor networks, and maintenance schedules. Traditional data warehouses and dashboards struggle to accommodate dynamic or unstructured inputs, e.g., PDF forms, sensor logs, etc., leaving decision-makers with incomplete or stale insights. The AFRL Earth 616 STRATFI program, led by SIMBA Chain, has successfully integrated data from disparate sources, including OEMs and Tier 1 suppliers, along with public data sources, including news, social media, public supply chain information and supplier provenance trails that identify suppliers’ overall risk. Earth 616 has also leveraged AI and semantic web technologies to structure unstructured data, enabling automated classification; with ontologies defining the domain model including entities, relationships, and constraints and synthetic Ontology Design Patterns (ODPs), to provide reusable templates capable of adapting to new contexts. This STTR Phase I, called Omniscient, or “all knowing”, provides a fabric to connect AI knowledge and data, making AI usable (by machines and humans) and scalable within Earth 616. Omniscient focuses on two key areas that build on this prior work and provide significant benefits such as common APIs, common knowledge patterns, scalability and simple user interaction to the use of Artificial Intelligence (AI) and Large Language Models (LLMs): Knowledge-Driven Data Fabric that provides an overarching architecture to host the various AI services to create a resilient, reusable, scalable layer for common and efficient machine to machine interaction for scaling Earth 616’s general strategy towards AI.Graph Retrieval-Augmented Generation (RAG) for intelligent user interaction that provides a backend API-driven agentic service that uses LLMs to convert from a simple natural language question to a Graph database query for the retrieval of machine processable results, with the agent passing the results to the LLM to convert back to a human readable format. The agentic approach should also be capable of selecting the appropriate source to ascertain the relevant answer, e.g., it could use the Graph database for a knowledge base to retrieve relevant facts, vectors for measuring difference between data points and LLMs for more responses that require more general knowledge.