SIGBOT using Natural Language Processing

Chatbots are rapidly becoming indispensable tools for improving student engagement and learning outcomes, especially in the context of personalized educational support. This project focuses on developing a subject-specific chatbot powered by Natural Language Processing (NLP), designed to align with academic curriculum and provide contextaware tutoring. The chatbot pipeline begins with PDF content extraction using PyMuPDF, with Tesseract OCR applied to scanned documents. Extracted text and images are structured into a knowledge base, where content is divided into overlapping chunks and linked with metadata such as file names, page numbers, and figures. For information retrieval, SIGBOT combines semantic similarity search through Sentence Transformers (all-MiniLM-L6-v2) with TF-IDF as a fallback, ensuring both meaning-based and keyword-based query handling. Answer generation is powered by a text-to-text model (FLAN-T5), which rephrases retrieved content into student-friendly explanations. Additionally, the system is integrated with the Ollama framework, enabling efficient local deployment and management of large language models (LLMs) for faster and privacy-preserving inference.