Trisurya: A Local Language-Powered Omnichannel Chatbot as a Solution for the Enhancement of E-Govemment Quality and Accessibility of Public Services in Indonesia
Arya Raditya Kusuma, Carleano Ravelza Wongso, Darren Aldrich, Fariz Darari · 2024
E-government initiatives have evolved in various forms in Indonesia over the past several decades. However, the services remain far from optimal; information sources are still fragmented and inaccessible to people who are not fluent in Indonesian. Thus, Trisurya is introduced as a chatbot capable of providing information from government and public service officials in local languages, in addition to Indonesian. Our system is built on Large Language Models (LLMs), specifically GPT API, enhanced by Retrieval-Augmented Generation (RAG) and leveraged by graph (Neo4j) and relational (PostgreSQL) databases to ensure the accuracy of the information provided. This study also compares two versions of the GPT API, GPT-4 and GPT-4o, to determine which version performs better as the LLM backend of the Trisurya system. While both offer similar performance, GPT-4o is more cost-effective. The evaluation results of Trisurya demonstrate its superiority over other publicly available and widely used chatbots, i.e., the vanilla ChatGPT-4o, ChatGPT-3.5, Gemini-1.0-Pro, and Claude-3-Sonnet. Based on the result, Trisurya demonstrated clear superiority during quantitative and qualitative assessments by generating accurate and high-quality responses in Javanese, Sundanese, and Balinese. This research represents a breakthrough in technological innovation by enhancing the efficiency of public services and revolutionizing the management and dissemination of government information in an inclusive manner through Trisurya's omnichannel chatbot.