KNOWL: Retrieval-augmented generation and hybrid knowledge-graph retrieval for more reliable enterprise LLM applications — Context and state of the art

Enterprise adoption of large language models (LLMs) is increasingly constrained by requirements for traceability, privacy, security, multilingual support, and domain-specific reliability. Retrieval-Augmented Generation (RAG) has emerged as a practical architecture for grounding LLM outputs in authoritative corpora, but baseline RAG pipelines often struggle with complex, multi-hop questions, long and heterogeneous PDF collections, and enterprise governance constraints. In parallel, knowledge graphs (KGs) and graph-based retrieval—popularized recently under the term GraphRAG—offer a complementary mechanism to model entities, documents, and their relationships, enabling retrieval and synthesis that can incorporate both local evidence and corpus-level structure. This paper presents the context and up-to-date state of the art for KNOWL, a project aimed at developing new RAG and hybrid (entity + document) graph solutions to improve LLM responses in enterprise settings, motivated by multilingual (English/Spanish) workflows for proposal and bid preparation in response to Requests for Proposals (RFPs). The focus of this first paper is deliberately non-evaluative: it synthesizes recent academic literature on RAG, GraphRAG, KG–LLM integration, multilingual retrieval and grounding, and PDF-centric document processing. It also reviews governance considerations centered on the EU Artificial Intelligence Act and GDPR and summarizes industry adoption patterns and operational risks from reputable reports. Finally, it articulates research gaps and motivates the hypotheses and evaluation direction that will be addressed in a subsequent methods-and-results paper.