Devon Messecar, United States Space Force
https://doi.org/10.64861/PORT5599
Keywords: agentic, AI, data broker, data federation, agents, large language model, data feeds, data access, data discoverability, UDL
Abstract:
Organizations invest heavily in centralized data platforms and still struggle to make their holdings discoverable, trustworthy, and reusable. We argue that this stems from conflating two distinct problems: the discoverability of data and the analysis of data. Before an analyst can advance an algorithm concerning a specific threat in a region, they must first find, understand, and access the datasets related to that region. This paper introduces the AI Librarian, an agentic approach to that foundational discoverability problem, and describes its reference implementation, the Agentic Data Broker (ADB). The AI Librarian is not another analytics platform. It is the card catalog for the enterprise, an ecosystem of specialized autonomous agents that mimic an efficient data team and separate the data plane from the analytics plane. Agents monitor designated environments for new sources, profile and interpret each one using ontology-grounded extraction, place it in the broader data landscape as a governed data product in a shared knowledge graph, and answer natural-language questions with path-based retrieval, so that every claim in an answer traces to a confidence score, an asserting agent, and a provenance record. We demonstrate the system end to end on satellite catalog data, propagating orbits from two-line element sets, deriving regional coverage, and answering multi-source questions with full provenance. We close with limitations and directions for future work.
Date of Conference: September 15-18, 2026
Track: AI for SDA Applications