Development of an Offline Interactive Chemistry Tutor Using Generative AI and Large Language Model

Abstract

This research presents the development of ChemGenAI, an offline, interactive chemistry tutor powered by a large language model (LLM) and cheminformatics tools, designed to enhance digital chemistry education in environments with limited or no internet connectivity. The primary aim is to create a locally deployable platform that supports real-time natural language interaction, molecular structure visualization, and periodic table exploration without relying on cloud-based services. ChemGenAI integrates the pre-trained Mistral-7B-Instruct model executed through Ollama for offline language processing and RDKit for molecular parsing and rendering. The system features a user-friendly graphical interface with three main modules: Chemistry Tutor, Molecule Draw, and Periodic Table. Results show that ChemGenAI accurately responds to chemistry questions, generates and visualizes 2D and 3D molecular structures from SMILES input, and presents detailed elemental data interactively. The interface design follows established usability heuristics to ensure intuitive navigation and engagement. The findings demonstrate that ChemGenAI offers an effective and accessible solution for chemistry instruction, particularly in classrooms where cloud-based systems are not feasible due to limited infrastructure or institutional that restrict external data access. The study concludes that offline AI tools like ChemGenAI can play a significant role in supporting inclusive, adaptable, and technology-enhanced science education.