No-Code RAG Ecosystems as an Alternative to Linear English Learning in Higher Education

Clavijo-Céspedes R.H.; Cobeñas-Chanduvi M.A.; Casariego-Neyra C.M.; Garay-Silupú E. · CEUR Workshop Proceedings (2026)

Resumen

Developing adaptive educational software typically demands significant technical and financial investment, limiting access for many institutions. This paper evaluates the effectiveness of No-Code ecosystems based on Retrieval-Augmented Generation (RAG) compared to traditional linear platforms and baseline generative AI models for English language learning in higher education. By configuring a “smart tutor” using Gemini Gems and NotebookLM, we demonstrate that explicit prompt engineering combined with strict document context injection creates a highly divergent, interactive, and zero-cost educational environment. To rigorously assess this approach, we conducted a three-way comparative evaluation utilizing a panel of five expert human evaluators. Our empirical results reveal that the proposed No-Code RAG ecosystem substantially outperforms both traditional rules-based platforms and unconstrained zero-shot Large Language Models. Specifically, the integrated system achieved an average score of 4.29 out of 5 in Factual Grounding, effectively eliminating the thematic hallucinations and domain mismatches that plagued the baseline generic model (which scored only 2.04/5). Furthermore, the No-Code tutor maintained a superior pedagogical tone (4.26/5) and fostered greater conversational engagement (4.19/5) by substituting direct answers with Socratic scaffolding. Ultimately, this research provides a fully replicable methodology to democratize advanced, hallucination-free educational technology in university settings without requiring custom software engineering.

Instituciones: Universidad Tecnológica del Perú; Universidad César Vallejo