Agentic AI Framework Using Retrieval Augmented Generation for Intelligent Academic Question Answering in Higher Education
Universities require fast, consistent, and traceable academic information services, yet direct use of large language models still poses risks of hallucination, outdated knowledge, and unsupported answers. This study developed an Agentic Artificial Intelligence framework based on Retrieval-Augmented Generation for intelligent academic question answering. The study employed Design Science Research comprising problem identification, requirements analysis, architecture design, prototype implementation, retrieval benchmarking, and evaluation. The framework integrated an agent orchestrator, intent routing, query rewriting, hybrid retrieval, reranking, memory, policy guardrails, a generator, and a verifier to check grounding and answer sources. Controlled testing used 12 synthetic academic-policy documents and 216 questions covering course registration, payments, academic leave, graduation, theses, MBKM, scholarships, library services, LMS, transcripts, academic calendars, and administration. Baseline RAG achieved Hit@1 of 80.6%, Recall@3 of 93.5%, and MRR of 87.6%, whereas Agentic RAG achieved 96.3%, 99.5%, and 97.8%, respectively. The improvement mainly resulted from intent-aware query expansion and combined lexical-character retrieval, which improved handling of paraphrases and overlapping service terminology. The proposed framework offers a more structured, transparent, and auditable implementation pattern for AI-based academic services. Field evaluation using an institution’s official corpus remains necessary before operational deployment.
Authors
- Eko Aziz Apriadi
- Fathoni Dwi Atmoko
Institutions
- Universitas Indonesia Mandiri (ID)
Publication Details
- Journal
- Jurnal JEETech
- Published
- 2026-11-16
- DOI
- https://doi.org/10.32492/jeetech.v7i2.7214
- Primary Topic
- Topic Modeling
- Type
- article
- Field-Weighted Citation Impact
- 0.00