AI in use for the analysis of company data

It is well known that large language models in artificial intelligence have problems interpreting structured data. This is because LLMs are mainly trained on unstructured text such as books, articles and web pages. Structured data, such as company data, often has implicit context conveyed by its structure itself: a table with column headings such as "Customer", "Product", "Quantity" and "Sales" provides information that must be understood by its position in the table and a calculation rule. Models that process natural language must learn to understand these implicit structures

Retrieval Augmented Generation (RAG) systems deliver excellent results in this respect. They combine the capabilities of retrieval systems with those of generative LLM models. The retrieval part of the RAG system can be used to precisely retrieve relevant data from large structured data sets. This can be done through search algorithms that can identify and extract specific entries or sections of data, whether from a table, database or other structured data set. Once the retrieval component has extracted relevant data, the generative component of the RAG system can interpret this data and provide detailed analysis. In the case of the question "Which three products have had the highest sales growth in the last five years and why?", the retrieval system first searches for the sales figures for all products, then identifies the ones that have grown the most and transfers the data to the LMM. This then provides the comprehensive answers. Curious? We'll show you how in our workshop.