Event language
UI language
Most RAG demos break down when they meet real-world constraints: complex tables, BI outputs, proprietary PDFs, security requirements, and cost limits.<br>In this talk, we share a production case study from Pretagov on how we customised the open-source Onyx AI stack to build reliable, evidence-backed AI assistants for government and enterprise environments.<br>Rather than focusing on generic chatbots, we’ll walk through the practical engineering work required to make RAG systems usable in regulated settings:<br><ul><li>Extending RAG beyond plain text to handle tables, reports, and structured data</li><li>Building custom ingestion pipelines for CMS platforms, BI tools, and proprietary documents</li><li>Enforcing citation, traceability, and governance over source material</li><li>Optimising cost and performance for real deployment, not demos</li></ul>This session is aimed at teams who need <strong>bespoke AI systems</strong> rather than off-the-shelf tools, and want to understand what it actually takes to run FOSS-based RAG in production.
