Information Retrieval Using Retrieval-Augmented Generation on PIRLS Documents

A comparative evaluation of baseline and agentic pipelines using open-weight LLMs and DeepEval metrics
Persadha, Widianto Sibberns, Heiko Tariq, Mohammad S. Wietzorek, Bettina
(2026)
This study explores how Retrieval-Augmented Generation (RAG) and locally hosted open-source language models can improve access to the information held in IEA's documentation, using PlRLS 2021 reports, technical documentation, questionnaires, and user guides as the test corpus.