Army TRUST-LLMOps Enabled Acquisition Requirement Package

This Phase II project solves some key challenges in the Army acquisition requirement package (ARP) including: 1) the manual process of developing acquisition documents according to requirements and templates, 2) the manual process of rewriting/reusing existing documents, 3) the manual process of compliance check and consistent updates across documents, and 4) the lack of access to the most relevant and accurate historical documents and prior examples. Large language models (LLMs) play a crucial role for the Army’s ARP, but their tendency to hallucinate and lack of transparency pose additional challenges to trusted applications.Expression proposes to address these challenges by our innovative solution, Trustworthy, Reliable, Unassailable, Scalable, and Transparent LLM Operations (TRUST-LLMOps), embedded in Visualize, Observe, and React (VOR) artificial intelligence (AI), VOR.AI, an LLM-powered workflow management and accelerator software tool. The Phase II project conducts research and development on the TRUST-LLMOps solution to address reliability and transparency, the two critical aspects of trusted LLM-empowered applications. In addition, it builds a functional prototype of VOR.AI which integrates TRUST-LLMOps with and optimizes it for the Army’s ARP.The anticipated major outcomes of this Phase II project include: i) the TRUST-LLMOps solution framework, algorithms, and engineering processes that ensure the reliability and transparency of the VOR.AI product across a broad set of application areas and customer segments; ii) a functional prototype of the first VOR.AI product tailored to the Army acquisition package; and iii) automated, trusted, and human-in-the-loop ARP workflows with significantly reduced solicitation cycle time and operational cost, as well as improved accuracy and consistency in solicitation documents.