Empowering Air Force with Large Language Model-powered Retrieval Augmented Generation

While Large Language Models (LLMs) have captured the imagination of many who hope to process a lot of unstructured, or free-form text, they suffer from several issues.  First, they often “hallucinate”, which means they respond inaccurately.  Second, they require a lot of data and time to be able to process domain-specific information.  In a rapidly evolving threat environment where novel information is appearing for the first time, both of these challenges can be fatal to mission execution.  Kyndi has solved the problem of hallucinations with a Retrieval Augmented Generation (RAG) system that can eliminate hallucinations, even on new and novel data.  The result is an AI platform that is accessible to non-technical people and which can find answers in one-click, but does not require labeled training data or skilled AI engineers, which equates to operational effectiveness (95%+ reduction in analysis time) in weeks, not months or quarters. Kyndi leverages some of the same underlying technologies as ChatGPT and other LLMs and can therefore find answers to questions, but Kyndi returns accurate and trustworthy answers verbatim instead of generating a new and novel response, which will be wrong an appreciable percentage of the time.  Kyndi will interview relevant NASIC stakeholders to identify NASIC-specific domain information to add to existing large language models to enable processing relevant intelligence and subsequent display on a Common Operating Picture (COP). ???????