Lettria Secures €4M to Build the Next Generation of AI for Document Intelligence

Lettria has raised €4 million in funding to lead an ambitious R&D program focused on hybrid AI for document intelligence. The initiative, LettRAGraph, is supported by France 2030 and conducted in partnership with Eurecom, a leading academic research center in knowledge engineering and natural language processing.

The goal: provide reliable, production-ready AI to industries where precision, traceability and compliance are non-negotiable — such as insurance, life science, finance and healthcare.

Rethinking RAG for Regulated Workflows

Traditional RAG (Retrieval-Augmented Generation) models, based on vector search alone, are insufficient when applied to complex enterprise use cases. They lack control, transparency and often fail to meet legal and operational standards.

Lettria is taking a different approach: combining LLMs with symbolic reasoning via knowledge graphs and domain ontologies. The objective is to build structured, explainable systems that can be deployed securely and integrated into real document workflows.

Three Technology Pillars

1. Multimodal Document Understanding

Handles diverse document formats — including scanned PDFs, tables, structured layouts and multilingual texts — to extract usable information from content that legacy NLP pipelines fail to process.

2. Text-to-Graph Structuring

Automatically extracts key entities, events, and relationships and maps them into a knowledge graph aligned with the client’s domain. This ensures that AI outputs are interpretable and consistent with internal business logic.

3. Hybrid LLM + KG Reasoning

Combines vector embeddings and symbolic graph-based logic to power applications such as semantic search, decision support, and audit-friendly document analysis — all grounded in structured knowledge.

Two Flagship Products Under Development

Product 1: Supercharge Your Knowledge Graph

This module enables companies to build, enrich and maintain business-specific knowledge graphs from unstructured data.

Key features:

• Automatic ontology creation and alignment using internal terminology
• Extraction of domain-specific relationships across heterogeneous documents
• Integration with leading graph databases (Neo4j, Ontotext, Amazon Neptune)
• Compatibility with multilingual corpora and regulatory taxonomies

Use case examples:

• Insurance: build policy clause graphs for comparison and audit
• Legal: extract precedent relationships from case law
• Finance: track entities and risk exposure across multi-source documents

Target users include organizations already using graph databases and looking to increase automation, quality and completeness.

Product 2: GraphRAG – Talk to Your Data

A conversational interface powered by hybrid retrieval: vector embeddings for flexibility, symbolic graphs for control.

Key features:

• Natural language query interface connected to enterprise knowledge
• Guardrails to reduce hallucinations and ensure traceability
• Role-based access control and output filtering
• Custom answer routing based on user profile or regulatory context

Use case examples:

• Compliance teams querying structured document repositories
• Legal analysts searching for obligations and exceptions across contracts
• Claims analysts asking questions across structured and unstructured reports

Unlike generic chatbots, GraphRAG uses curated knowledge and business logic to generate responses that are both verifiable and aligned with internal rules.

Timeline and Access

• Program duration: 36 months (started April 2024)
• First features available for pilot testing: from Q4 2025
• Early access is available through Lettria’s pilot program, including onboarding support and custom evaluation

👉 To evaluate how these modules can support your internal workflows, contact our team to schedule a live demo.