A Hybrid RAG for Answering Questions about Technical Documentation from Github Projects

Abstract

Software technical knowledge is often fragmented across PDFs, code repositories, and Web documentation. This work presents a hybrid retrieval-augmented generation system for answering design-focused questions over heterogeneous technical sources. The prototype combines semantic search across multiple knowledge sources with source preservation and attribution to reduce hallucinations. An initial evaluation on software-design questions indicates accurate answers and effective source attribution, particularly when information must be combined across sources.

Publication
Proceedings of the Argentine Symposium on Software Engineering (ASSE), 55th JAIIO 2026, La Plata, Argentina