Celery (Apium graveolens) has a high economic value because it is widely used as a food ingredient, cooking spice, and even herbal medicine. However, sometimes the cultivation of this plant faces various challenges, especially diseases and pests that can reduce its quality and yield. This study aims to develop an expert system based on Natural Language Processing (NLP) with a Retrieval-Augmented Generation (RAG) approach to help diagnose plant diseases quickly and accurately. This research was conducted using the Extreme Programming development method. The research methodology includes identifying celery cultivation problems, collecting data from experts regarding various symptoms, celery diseases, and effective treatments, and developing RAG using Transformer-based embedding techniques that have proven effective in capturing context and relationships between words. This system uses forward chaining reasoning to ensure that the solutions provided are generated through an inference process from initial facts such as symptoms to reach appropriate conclusions. These results show that this system is able to identify disease symptoms from user text input with 97.14% accuracy and provide relevant solutions. Equipped with speech-to-text, text-to-speech features, as well as the ability to copy answer results and delete conversation history to make it easier for users.