Generative Text Engine for Form Completion

The U.S. Navy requires that ships have all their systems ready to perform specific missions as tasked by their operational commanders. These highly complex systems require preventive and restorative maintenance to be performed regularly to maintain optimal performance. Data accuracy, in the form of documented maintenance actions, is critical to deriving consistent meaning from the aggregated pool of input. This Phase I project will address how to automate a curation process, incorporating Machine Learning tools, specifically Natural Language Processing and Large Language Models (LLMs), to increase the accuracy of end-user input in operational environments. The project has four objectives: 1) Incorporate specialized language/jargon without having to train the LLM, 2) Run speech-to-text functionality on appropriate mobile hardware, taking the place of manual methods such as the “wheel book,” 3) Run a suitable LLM on a local server, and 4) Implement a prompting workflow that drives form completion. Phase I will consist of researching these questions, identifying practical solutions, and assessing these solutions through a combination of technical analysis and practical testing to determine which solutions are the most suitable. Based on our experience with integrated systems involving LLMs, we will derive solutions for each objective.