Automating Information Retrieval with Semantic Data Processing

This report proposes to consider the possibility of automating the search for the necessary information in a corporate database that may contain data of various formats. To implement such a process, it is necessary to use large language models (LLM), but to generate high-quality responses, it is necessary to further train the models using Retrieval-Augmented Generation (RAG) technology, which creates vector databases based on corporate data. These repositories can be constantly updated to assist generative Artificial Intelligence (AI) in providing optimal contextual and timely responses, in addition to connecting the semantic ranking of the results obtained for enhanced accuracy. The work also proposes to evaluate the work of language models using known metrics. This is a result of studying which the problem of the occurrence of errors of the 1st and 2nd kind when generating a response by the LLM model is revealed. Consequently, it has been determined that no universal assessment of models exists.