- editor.aipublications@gmail.com
- Track Your Paper
- Contact Us
- ISSN: 2456-2319
Kanika Chalana , Karthick S , Syed Usman
International Journal of Electrical, Electronics and Computers (IJECC), Vol-11,Issue-1 (Special Issue-NCASTEM), January - February 2026, Pages 66-73, 10.22161/eec.ncastem.2025-9
Download | Downloads : 3 | Total View : 404
Article Info: Received: 20 Nov 2025; Accepted: 03 Feb 2025; Date of Publication: 10 Feb 2026
In recent years, the integration of Artificial Intelligence (AI) into educational technology has shown promising potential in enhancing student engagement and access to learning resources. This paper presents the development of a domain-specific chatbot trained on VTU-prescribed textbooks, with an initial focus on the major subjects. The chatbot is designed to serve as an intelligent academic assistant, capable of answering student queries in natural language while adhering strictly to the prescribed curriculum. The chatbot employs a Retrieval-Augmented Generation (RAG) architecture that combines the semantic understanding capabilities of Large Language Models (LLMs) with the precision of document retrieval systems. The core model used is Mistral, accessed via the ChatOllama interface, chosen for its balanced performance in generating informative and context-aware responses. For embedding and indexing textual data from the textbook, the system utilizes HuggingFace’s sentence-transformers (all-MiniLM-L6-v2) to generate dense vector representations, which are stored and queried using the Chroma vector database. To ensure relevance and reliability, the chatbot limits its response generation to the top-k most semantically similar document chunks retrieved from the textbook. Responses are presented in a structured Markdown format, including an answer section, supporting evidence, and source references. The chatbot is deployed using Streamlit, offering an interactive web interface where users can engage in real-time conversations, ask questions related to syllabus, and receive syllabus-aligned responses. This research demonstrates the feasibility and effectiveness of constructing specialized academic chatbots tailored to institutional curricula. By narrowing the knowledge domain and grounding the model in verified academic sources, the system avoids hallucinations and enhances trustworthiness, making it a practical supplement to traditional learning. Future work will expand this framework to other subjects within the VTU curriculum and incorporate automated evaluation techniques to measure response accuracy and educational impact.
