Development of BAPOLAIC: AI chatbot for optical character recognition based-document extraction and voice assistant

Abstract

Conventional chatbots often lack integrated functionalities for complex academic tasks, such as multi-format document handling and multimodal interaction. This paper presents the design, implementation, and performance evaluation of BAPOLAIC, a web-based, multimodal AI assistant developed to address this gap. The system architecture integrates optical character recognition (OCR), a dual-strategy natural language processing (NLP) module, and voice assistance, all orchestrated by the Gemini API. Quantitative evaluation confirmed high performance: the OCR module achieved a 98.69% average accuracy, and the retrieval-based NLP path correctly handled 90% of test queries. Furthermore, the API integration demonstrated exceptional efficiency with a median latency as low as 0.06 ms. Task-based evaluations validated BAPOLAIC's effectiveness in performing intelligent functions like summarization and content-based Q&A, with a superior capacity for handling up to 10 consecutive documents. The results validate BAPOLAIC as a successful proof-of-concept for a specialized academic tool, providing a framework for integrating multiple AI technologies to enhance educational productivity.