AI-Assisted Letter Validation and Automated Response Generation for Official Letters

Official letters often need to be checked for required information before they can be processed further. This checking is usually done manually and can take time, especially when the letters are received as scanned documents or images. This paper presents an AI-assisted system developed to automate the validation of such letters and generate responses for incomplete submissions. The system accepts letters in PDF, PNG, and JPEG formats and uses Optical Character Recognition (OCR) to extract their text. The extracted text and document details are stored in a database and passed to the validation stage. The implemented format validator checks for required elements such as the date, subject, salutation, sender, and recipient. For content validation, the system is designed to use a domain-specific knowledge base with Retrieval-Augmented Generation (RAG) and a Large Language Model (LLM) to identify missing or incorrect information. The validation findings are then used to generate a response describing the information that needs to be provided or corrected. The system is implemented using Java and Spring Boot with separate components for document processing, OCR, database management, and validation. Synthetic MRSAC-related data is used during development of the AI-assisted components because the actual organizational requirements are confidential.