Generative Artificial Intelligence for Course and Content Creation and Conversion (GenAI4C)

Team Sonalysts proposes developing the Advanced Training and Learning AI System (ATLAS), a Retrieval Augmented Generation (RAG)-based Large Language Model (LLM) tool that increases the speed, quality, and consistency with which instructional material (including required documentation) is updated or created. ATLAS employs a Modular Open Systems Architecture. This frees the Government from restrictive licenses and promotes vendor competition. Moreover, this approach enables incremental development and speeds the process of deploying functional tools. ATLAS prioritizes the use of open-source components to reduce costs and foster community involvement. Its well-defined public Application Programming Interfaces ensure adherence to well-documented industry standards that promote extensibility and competition. ATLAS is designed to interface naturally with the USMC training infrastructure. It employs a Moodle-based Technology Readiness Level (TRL) 7 User Interface built on the open-source Open Web User Interface that is authorized to process Controlled Unclassified Information (CUI) and deployable to classified environments. This will foster transition by reducing barriers to obtaining an Authority to Operate and allowing the system to work with public domain, CUI, and classified information. Internally, ATLAS is architected to support the full range of USMC instructional design tasks, including curriculum refresh and novel development. To illustrate this, consider a few of ATLASĺs service modules. The ATLAS LLM Orchestration Module manages interaction with the chosen LLM(s) and implements the RAG pipeline, freeing ATLAS from dependency on any given LLM approach. The Multimodal Generation Module generates images, videos, and other multimedia assets to enhance instructional materials, allowing ATLAS to provide the full range of content needed in modern learning environments. The Knowledge Base Module stores and manages instructional design knowledge, learning standards, and various technical sources, enabling curriculum designers to stay abreast of USMC tactics, techniques, procedures, and technology. ATLAS employs an event-driven architecture with comprehensive logging, monitoring, and tracing. The event-driven architecture ensures content currency. For example, suppose the source material within the LLM knowledge base is updated. In that case, it automatically triggers a workflow for the LLM and designer to evaluate whether the associated curriculum material should be updated. Comprehensive logging, monitoring, and tracing provide a range of benefits, including faster load times, smoother user experiences, and enhanced scalability and responsiveness.