PDF-Document Chatbot Responses using Large Language Models to Enable Smart City Engagement

Abstrak

Traditional documents, including Rencana Pembangunan Jangka Menengah Daerah (RPJMD), Strategic Plans (Renstra), and e-masterplans, have undergone a remarkable transformation, evolving from their conventional printed formats to the dynamic realm of electronic versions. While this shift holds the promise of enhanced accessibility and convenience for the public, the full potential of these resources remains unrealized due to inherent challenges. On the other hand, a Generative AI approach is employed for the creation of an intelligent chatbot. Our primary contribution lies in the PDF-Document Chatbot Response utilizing Large Language Models (LLMs) GPT 3.5 Turbo from OpenAI, aimed at fostering engagement within Smart City. The dataset consists of Masterplan documents for Smart City development in Yogyakarta City, presented in PDF format and employing the Indonesian language. This research leverages the Large Language Models (LLMs) GPT-3.5 Turbo from OpenAI, in conjunction with user input and prompts. The development process for crafting this chatbot utilizes the LangChain Framework and Pinecone for storing vector embeddings. The results underscore the chatbot's capability to generate coherent responses closely aligned with the context found within the PDF document.