Google Gemini-Based Service Chatbot for Indomaret Pekanbaru

Abstract

Modern retail customer service requires fast, consistent, and accessible information delivery. This study implements a text-based chatbot prototype for Indomaret Pekanbaru services using Google Gemini and semantic similarity retrieval. The dataset uses a question-answering structure that contains service contexts, user questions, and expected answers related to operating hours, payment methods, promotions, digital top-up services, bill payments, package delivery, and branch information. User input is normalized through preprocessing, converted into sentence embeddings with Sentence Transformer, and compared with dataset questions using cosine similarity to select the most relevant context. The selected context and user question are then sent to the Gemini API to generate a contextual response. Functional testing used five service-question scenarios. The chatbot produced appropriate answers in three scenarios and partially appropriate answers in two scenarios. The main limitation appears when requested information is too specific or unavailable in the dataset. The results indicate that semantic retrieval and generative language models can support flexible customer service automation, but dataset expansion and response validation remain necessary.