Použití umělé inteligence pro extrakci textu z dokumentů a jeho integraci do aplikace pro úvěrovou analýzu

This thesis explores the use of artificial intelligence for automating data extraction from unstructured financial documents, with a focus on credit analysis reports in the banking sector. Credit analysts routinely produce detailed reports when evaluating loan applications, but the information contained in these documents is rarely reused due to a lack of structured integration. To address this inefficiency, the thesis develops and validates a proof-of-concept pipeline that extracts both structured and unstructured data from PDF documents and integrates it into a centralized SQL database that powers a credit analysis application. The solution combines Azure OpenAI GPT-4 Omni for processing free-text narratives with Azure AI Document Intelligence for extracting structured data. The thesis follows the CRISP-DM methodology and is contextualized using the three-axis maturity framework, which evaluates the organization’s readiness across analytics, data, and IT dimensions. The implementation is iteratively refined based on feedback from credit analysts, ensuring alignment with real-world workflows. The results demonstrate the feasibility and business value of integrating cognitive AI into document-heavy processes and lay the foundation for future expansion of AI capabilities within the bank’s credit risk management systems.