https://www.mdu.se/

Large organizations today have an extensive amount of unstructured data that needs to be processed, such as PDF documents containing free text, tables, and values expressed in different formats or languages. At Ericsson, information from Request for Quotation documents is currently extracted manually, which is time-consuming, costly and prone to error. This thesis investigates how existing AI-based text management tools can be combined to automate this process, and whether Microsoft Fabric can fulfill the company’s needs for extracting and normalizing information from unstructured Request for Quotation documents.  The work focuses on vehicle leasing Request for Quotation PDF files and presents the development of a prototype workflow in Microsoft Fabric. The prototype ingests PDF files from SharePoint, processes them through a Python notebook using Fabric AI functions, extracts selected vehicle and pricing values, normalizes the result into standardized English terms, and exports the structured output into an Excel file. To include a human-in-the-loop approach, a Needs Review column is included for values that require manual validation.  The results show that Microsoft Fabric can be used to orchestrate a document processing workflow and that the prototype performs well on text-based Request for Quotation documents with varying formats and languages. However, there are limitations, especially for image-based documents where the extraction produced unreliable results. The thesis contributes a practical prototype for automated Request for Quotation information extraction in an enterprise context.