Customer stories
From 20% to 90% Accuracy: How Orbis Rebuilt RAG App with MCP
Orbis Holding set out to build a natural language Q&A system over a massive, real world surveillance dataset. What they got instead was a brittle LLM to Cypher pipeline that looked correct, failed silently, and collapsed under encoded values, multilingual data, and multi hop relationships.
In this success story, learn how they rebuilt their RAG architecture using Memgraph and an MCP-based refinement loop. By inspecting live schema and values at runtime and iteratively correcting queries, they transformed an unreliable prototype into a production-ready system that delivers accurate answers at scale, even across 100 million interconnected nodes.
What you’ll learn:
In this success story, learn how they rebuilt their RAG architecture using Memgraph and an MCP-based refinement loop. By inspecting live schema and values at runtime and iteratively correcting queries, they transformed an unreliable prototype into a production-ready system that delivers accurate answers at scale, even across 100 million interconnected nodes.
What you’ll learn:
- Why traditional one-shot LLM-to-Cypher pipelines fail on messy, real world graph data
- How MCP enables adaptive query refinement instead of silent empty results
- How Orbis reliably queries encoded, abbreviated, and translated multilingual properties without manual fixes
- How Memgraph handles low latency, multi-hop traversal at 100M node scale with consistent accuracy
