Enterprise RAG in Liferay AI Hub: Grounded, Governed, and Permission-Aware
Presenters:
- Feliphe Costa Marinho, Sr Software Engineer, Liferay
- Jose Abelenda, Team Lead, Liferay
Retrieval-Augmented Generation is easy to demo and hard to ship to an enterprise. The moment real users and real content are involved, three questions decide whether RAG is production-grade: Is the answer grounded in your content? Is the pipeline governed against misuse? And does it respect who is allowed to see what? This deep dive walks through how Liferay AI Hub answers all three, using LangChain4j as the orchestration backbone. We’ll trace a single chat turn end to end — from prompt to grounded answer — and stop at each architectural layer along the way. On grounding, we’ll look at how AI Hub retrieves semantically against an embedding index, returns scored, highlighted fragments rather than whole documents, and discards low-relevance chunks before they ever reach the context window — plus how it federates to a site’s own search API and Search Blueprints for hybrid retrieval. On permission-awareness, we’ll show how retrieval executes under the end user’s own identity and access token, so RAG can never surface content a user couldn’t otherwise reach — and how every retrieval is published for observability and audit. You’ll leave with a concrete reference architecture for building RAG that you can actually put in front of enterprise users — and a clear-eyed view of the trade-offs we made to get there.