HybridRAG: A Practical LLM-based ChatBot Framework based on Pre-Generated Q&A over Raw Unstructured Documents

Researchers introduced HybridRAG, a novel framework that pre-generates a question-answer knowledge base from raw, unstructured PDF documents (including text, tables, and figures) using OCR and LLMs. At query time, it first retrieves from this pre-built QA bank, only generating answers on-the-fly as a fallback. Tests on OHRBench show it delivers higher answer quality and significantly lower latency than standard RAG, making it practical for high-volume, resource-constrained real-world applications.