Lettria Knowledge Studio

Understanding GraphRAG

Lettria is taking retrieval-augmented generation (RAG) to the next level, merging graph technology with the standard vector approach to create something better. Our goal is to help enterprises benefit from RAG technology without the risks of hallucinations or black-box responses.

Set up a demo ->

A new type of RAG: Merging knowledge graphs with vector databases

Goodbye, Hallucinations

Current large language models (LLMs) are held back by the limitations of vector databases. The most obvious of these limitations are seen when the machine produces “hallucinations”. To address this problem and improve LLM accuracy, the RAG approach has been helpful, but not sufficient given the drawbacks of vectors.

That’s why Lettria has developed Knowledge Graphs, providing more context and accuracy in order to stop hallucinations and restore trust in generative AI.

The best of both worlds

At Lettria, we’re delivering a better AI future thanks to the hybridization of the two RAG worlds. The result is faster, more accurate, and more contextually-aware AI.

Vectors provide a fast, efficient filtering system that narrows down the search space. Knowledge graphs then step in to understand relationships and offer rich, traceable context.

Hello, GraphRAG

Lettria has built its Knowledge Studio on the GraphRAG approach. By merging the contextual richness of knowledge graphs with the dynamic power of RAG tasks, we provide the environment that LLMs need to more accurately answer the pressing questions in your company.

The result? Precise, relevant, and insightful answers that capture the true essence of your business knowledge.

Improved accuracy, performance and scalability

With GraphRAG, the idea of “chatting with your data” becomes a reality, as data is transformed from a store of knowledge to an active partner.

Your unstructured data becomes usable and useful, ready to respond to day-to-day needs throughout your business.

Set up a demo ->

How Lettria does it

Your texts, documents, and data – structured and unstructured – are turned into a graph.

1. Document importation and parsing

Each data source is carefully cleaned and preprocessed in order to extract text chunks and store metadata.

Entity recognition and linking Recognition and Linking

The chunks are processed through Lettria’s natural language structuration API, identifying entities and relationships between them in order to produce a knowledge graph.

3. Embeddings and vector management

In parallel, the chunks are vectorized in order to maintain performance.

4. Database merging and reconciliation

The structured output from our API and the embeddings are stored in a single database, ready to power all of your RAG needs.

Start restoring your company’s trust in AI thanks to GraphRAG.

Set up your demo