Zpracování dokumentů využitím AI cloudových služeb

The aim of the work was to design a solution for selected document processing use cases using appropriate AI services on available cloud infrastructure in a banking environment. The focus was primarily on use cases involving translation, digitization, and extraction of specific attributes from ID documents. In the first implementation phase, the most suitable technologies for addressing these use cases were identified, compared, and selected, namely services from the Azure AI Services family: AI Translator and Document Intelligence. In subsequent phases, individual shortcomings of these services were identified and mitigated to meet the defined functional requirements. Some attention was also given to defining non-functional requirements and achieving them in connection with the secure and controllable operation of these solutions in a regulated banking environment. The final solution was built on the previously mentioned services together with the newly discovered Azure Content Understanding, which offers extended document processing capabilities required by some of the addressed use cases. The services were deployed and managed using Terraform IaC with the so-called Azure Verified Modules. For one of the solutions, custom wrapper code was built and deployed in an Azure serverless service to address certain identified limitations for digitization needs. The main contribution of the work focused both on the target solution artifacts (documentation, solution architecture, code, secure cloud service configuration) and on the enrichment through newly acquired knowledge, primarily for the needs of the given company and within the operating team. The main conclusions of the work, from a technological perspective, were primarily the need for further research and development of the Azure Foundry service as a newer offering with the potential to encompass end-to-end activities related to document processing and to follow the evolution of the Azure AI services product portfolio. From an operational perspective, the primary emphasis was on building CloudOps competence to enable effective use of AI cloud services and fulfillment of the company’s internal rules and requirements. From a decision-making perspective, the most important finding was the need to establish the right criteria for deciding between in-house development and waiting for a cloud-native solution.