BhashaRAG Privacy-Preserving Offline OCR and PDF-Based Multilingual Question Answering
BhashaRAG Privacy-Preserving Offline OCR and PDF-Based Multilingual Question Answering
Nayanapatruni Santosh Kumar
Department of Computer Applications, Visakha Institute of Engineering & Technology (Autonomous), Visakhapatnam, Andhra Pradesh, India
Master of Computer Applications, Jawaharlal Nehru Technological University–Gurajada Vizianagaram
Abstract
Educational institutions provide large volumes of information through websites, notices, PDF documents, and administrative offices. Locating simple information such as courses, fees, admissions, faculty details, and facilities can therefore become time-consuming for students and parents. BhashaRAG is an offline, privacy-preserving multilingual question-answering chatbot designed for educational institutions. The system combines a locally maintained JSON knowledge base with Retrieval-Augmented Generation (RAG), fuzzy matching, language detection, and Optical Character Recognition (OCR). Users can interact with the application in English, Telugu, or Hindi, while image-based inputs can be converted into text through the OCR module and passed through the same query-processing pipeline. The application uses Python and Flask for backend processing and HTML, CSS, and JavaScript for the web interface. Unlike cloud-dependent chatbot solutions, the developed system performs its core processing locally and does not require an active internet connection for retrieving institutional information. Functional testing of the implemented application covered multilingual interaction, course and fee retrieval, specific programme queries, OCR extraction, and interface controls. The reported OCR test extracted 643 characters from a sample image, and the tested retrieval functions returned course, fee, and programme-specific information from the local knowledge base. The results demonstrate the feasibility of combining offline retrieval, multilingual interaction, OCR, and fuzzy matching in a single educational information assistant.
Keywords: BhashaRAG, Retrieval-Augmented Generation, OCR, Multilingual Chatbot, Offline AI, Fuzzy Matching, Educational Information Retrieval, Privacy