Sponsored by military and industry organizations including the Central Florida Tech Grove, the Office of Naval Research (ONR), the Naval Air Warfare Center Training Systems Division (NAWCTSD), and the National Security Innovation Network (NSIN), the Ready Relevant Learning Challenge is a call to industry to identify AI-based solutions that could enhance and accelerate training development.

From initial interest by 61 companies, four finalists were selected to demonstrate their prototypes. Aptima captured first place for NAUTICAL, the Novel AI Utility for Training, Instruction, and Comprehensive Analytic Learning.

NAUTICAL is an AI platform that harnesses Generative AI and large language models (LLMs) to accelerate training development. It can take Instructional System Designers and Developers (ISDs) months to complete the process of task analysis, learning analysis, and media selection before new training content can be brought to life. Compounded by a shortage of ISD talent, this presents a significant backlog to developing training at the Navy’s speed of need. A web-based tool, NAUTICAL is designed to mirror the ISD workflow, performing as a ‘training expert’ alongside the human ISD to automate many of the analyses and outputs, yet preserving oversight to support transparency, explainability, and trust.

The inspiration behind NAUTICAL is to teach LLMs, which have been trained extensively on open-source information, to be an expert in training development. Rather than ISDs searching through materials or meeting with Navy subject matter experts (SMEs), which pulls them from their active-duty responsibilities for weeks at a time, NAUTICAL becomes a subject matter expert resource.

Using NAUTICAL’s “expert” prompt templates, an ISD fills in blank fields, like in a Mad Libs. These structured prompts ensure that the LLM’s responses are specific and accurate, with the templates able to be reused over and over.

To supplement the LLM’s knowledge with military data that’s not open-source, NAUTICAL uses Retrieval Augmented Generation (RAG) to ingest the necessary information. For a new job for example, it can source training doctrine or equipment manuals that have been uploaded to provide more precise, up-to-date answers.

To prove to the ISD why its responses are correct, NAUTICAL self-validates by including the LLM’s rationale and source data. This transparency and explainability provides the human confidence that the answers are grounded in fact and not a guess or hallucination. The ISD can accept, modify or reject the responses, which provides feedback to the model to learn from and continue to improve its answers.

NAUTICAL’s ability to generate answers in batch for hundreds or thousands of questions, provides repeatability throughout the ISD process, from task analysis to learning analysis to media selection. Rather than an ISD filling in answers in thousands of cells in massive spreadsheets, which can take months, NAUTICAL can develop the training requirement documents in a matter of days.

In the RRL Challenge, each of the final prototypes was issued a new task analysis to perform just 24 hours ahead of the final demonstration to prove its extensibility. NAUTICAL was able generate all the answers for the new analysis within hours, showing it can reduce the traditional ISD timeline from months to weeks or less.

Aptima, which developed and fine-tuned the NAUTICAL prototype in 8 weeks, showed how LLMs that are designed within a structured and guided workflow for trust, can be leveraged as an expert to assist humans, not to replace them.