ITEBIS PGRI Dewantara Academic Service Chatbot Based on IndoBERT and Retrieval-Augmented Generation (RAG)
Year: 2026
Type: article
Abstract: Manual administrative processes at ITEBIS PGRI Dewantara Jombang frequently cause delayed student feedback and overload staff with repetitive questions. To solve this issue, this research developed a 24-hour academic chatbot prototype that integrates a fine-tuned IndoBERT model within an extractive Retrieval-Augmented Generation (RAG) framework. The indobenchmark/indobert-base-p1 model was trained over five epochs using a dataset of 2,517 SQuAD-formatted academic question-and-answer pairs. For the architecture, the system employs Laravel to manage the frontend interface and relies on FastAPI as the backend to run the RAG pipeline. Upon evaluating the model using 504 validation pairs, the algorithm obtained an Exact Match (EM) accuracy of 71.32% and an F1-Score of 86.63%. In addition, functional black-box testing verified that the web application effectively extracts and displays accurate, real-time answers straight from a dynamic database. The primary novelty of this research is the implementation of RAG alongside IndoBERT specifically for Indonesian conversational systems, which eliminates the need to constantly retrain standard closed-book models. Ultimately, this paper offers practical end-to-end documentation on how to deploy a Natural Language Processing model into a fully operational web environment.
Source: Recursive Journal of Informatics
Authors: Syaiful Imron, Arbiati Faizah, Sugianto
Institution: Institut Teknologi dan Bisnis PGRI Dewantara Jombang
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
SDG: Quality education
Open Access status: bronze
OpenAlex ID: W7215061618