TransForm: Generative AI for Filling Forms

Manual form filling is an important step in military and business workflows. Forms that contain free-text fields pose a processing challenge because unexpected wording, phrasing, and jargon can hinder automated analytics downstream. A text-processing solution is needed that enables the form-processing software to interpret and revise the free-text fields. Charles River Analytics proposes to design and demonstrate TransForm, a system that uses generative artificial intelligence (AI) technology to revise form fields to use best practices and facilitate automated processing by downstream analytics. TransForm will employ advanced AI techniques including innovative large language model (LLM) fine-tuning and prompt-tuning approaches. Specific techniques include retrieval-augmented generation (RAG) and novel chain-of-thought reasoning methods. A well-designed human-machine interface supports the adaptation of TransForm to new terminologies, phraseology, and reasoning processes. TransForm is designed to be accessible through efficient human–machine interfaces from phones and tablets, and end users do not require any special computational or linguistic skill.