Vývoj webové aplikace využívající generativní AI pro predikci chemických reakcí

s part of this diploma thesis, a fully functional web-based tool for teaching chemistry has been developed, combining a generative AI model with an interactive virtual laboratory. A prediction model based on the phi3:mini language model and the Ollama tool was implemen- ted, capable of predicting the products of chemical reactions, including balanced equations, based on specified reactants and conditions (temperature). Unlike existing specialized solu- tions (Molecular Transformer, ReactionT5), the resulting model can predict the products of both organic and inorganic reactions, thanks to its integration with web search, which ensures the timeliness and relevance of information. The work also includes a web application simulating a laboratory environment, which allows users to mix chemicals in equipment of three different types (beaker, flask, test tube), heat 4 them using a burner, and create aqueous solutions. The application supports two modes – guest mode (quick access without registration) and logged-in user mode (saving environments, favorite chemicals). A key contribution is the storage of prediction results in a database, which reduces response time for repeated experiments from tens of seconds to milliseconds, making the system progressively faster. Automated and manual testing of all parts of the system confirmed the functionality of both the model and the entire application.