Meridian
AI Knowledge Base.

A production RAG system that lets 200 consultants query a decade of internal documents, methodologies, and case studies in plain English — and get accurate answers in under a second.

Making institutional knowledge instantly accessible.

Meridian's consultants were spending hours every week hunting through shared drives, Confluence pages, and email threads for answers that existed somewhere in the company's institutional knowledge. The information was there — it just wasn't findable.

We designed and built a RAG-powered knowledge base on top of their existing document library. Consultants now ask questions in plain English and get cited, accurate answers drawn directly from internal sources. The system handles 300+ queries a day with a 94% accuracy rate on their internal benchmark suite.

The project covered everything from chunk strategy and embedding model selection through to the chat UI and role-based access controls. We deployed to their existing AWS infrastructure and handed over full documentation on day one.

Building a retrieval system accurate enough for professional consulting use — where a hallucinated answer has real business consequences.

Accuracy isn't optional when the stakes are real.

Most RAG demos look impressive until they hallucinate. For a consulting firm where wrong answers erode client trust, we needed a system that knew what it didn't know — and said so clearly rather than confabulating.

We tackled this through a combination of hybrid search (dense + sparse retrieval), a reranking layer, and a strict citation requirement baked into the system prompt. Every answer the system returns includes the specific document and page it drew from. If the retrieval confidence falls below threshold, the system declines to answer rather than guessing.