INTEGRASI LLM DAN RAG DALAM PLATFORM CHATBOT TERPADU UNTUK LAYANAN INFORMASI AKADEMIK

Abstract

The utilization of Artificial Intelligence in academic information services is often hindered by the fragmentation of various Large Language Models (LLMs) and the models' lack of specific institutional knowledge. This study aims to design and build a unified web-based chatbot platform that integrates various AI service providers, including Google Gemini, Qwen, and models via OpenRouter (Llama, DeepSeek, Microsoft Phi), into a single interface. The application was developed using PHP with a secure architecture, utilizing HTML Purifier for output security and a responsive interface design based on Tailwind CSS. The key feature of this system is the implementation of the Retrieval-Augmented Generation (RAG) method, which enables the chatbot to provide accurate answers regarding specific information of Universitas Graha Karya (UGK) by referring to a local knowledge base, thereby minimizing AI hallucinations. Furthermore, the system is enhanced with Text-to-Speech (TTS), image generation, and video generation capabilities via the DashScope API. Test results demonstrate that this platform effectively improves the efficiency of academic information access by offering the flexibility to select AI models according to user needs and providing contextual, academically valid responses.