Enterprise proposal writing as a response to RFPs is a very grave time-consuming undertaking since you must read and understand a stack of documentation correctly to ensure that everything is per the requirements. Of course, Large Language Models (LLMs) are able to generate fluent text, but in the context of depending solely on them they tend to produce either hallucinatory text or fail to pinpoint context as they are ungrounded. This paper demonstrates a Retrieval-Augmented Generation (RAG) system, which automates the process of proposal drafting in an enterprise by consuming docs and extracting text, chunking text into semantically meaningful objects, generating dense vectors, and performing similarity-based retrieval to provide the LLM with a real-world context. When a system runs, it retrieves the pertinent elements of a vector database in real-time and inserts them into a structured prompt and the LLM then runs, enhancing factual accuracy and maintaining the context tight. It cooperates with large API-based structures and smaller local ones, which means that you can make comparisons on performance, the efficiency of context handling, and the ability to deploy anything. Our experiments demonstrate that relevance and reduction in hallucinations are increased with the addition of retrieval over and above generating text blindly, and it is a scalable, modular means of putting AI to work on writing enterprise proposals.
S et al. (Thu,) studied this question.