How Lettria Unlocked 20% Accuracy Gains with Qdrant and Neo4j
summary
Lettria builds a GraphRAG pipeline combining a document parser, an ontology-driven knowledge graph and vector storage in Qdrant and Neo4j and reports a 20% accuracy improvement measured by blind linguist annotations while keeping sub-200ms latency. Engineering includes a custom sync engine that snapshots Qdrant vectors, applies optimistic writes, rolls back or merges on Neo4j transaction failures, flattens payloads for compatibility, uses payload indexing and hot/cold tiers, and preserves lineage metadata for explainability — the system succeeds in production.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:12Challenges in regulated industries
- 3:47Limitations of classic RAG
- 5:10Lettria’s pipelines and assets
- 9:10Key partnerships Qdrant and Neo4j
- 13:30Performance gains and benefits
- 14:55Conclusion and future outlook
related talks
39mEnhancing RAG with Neo4j Knowledge Graph
36mGraphRAG Explained 🔥 Building Smarter LLMs
20mPractical GraphRAG: Making LLMs smarter with Knowledge Graphs
16mWhen Vectors Break Down: Graph-Based RAG for Dense Enterprise Knowledge
30mPerformant GraphRAG Improving Neo4j to Drive Autonomous AI ...
34mDon’t Get Lost in Vector Space
53mPractical GraphRAG - Making LLMs smarter with Knowledge Graphs
20mHybridRAG: A Fusion of Graph and Vector Retrieval