Skip to content
Research & Innovation

Our deeptech commitment

Letxbe invests heavily in R&D to build sovereign, frugal AI models specialized for regulated transactions.

DeepTech Label

Letxbe is DeepTech-labeled by BPI France, recognizing our approach based on scientific research and breakthrough AI innovation.

What this means

  • AI models developed in-house, not wrappers on existing APIs
  • Fundamental research in document processing and transactional understanding
  • R&D team from academia (PhDs, publications)
  • Innovation protected by intellectual property
BPI France Funding

Letxbe is supported by Bpifrance for innovation. This funding validates our technological approach and our ability to create value in the French market.

  • Innovation funding for our transactional AI engine development
  • Validation by independent experts of technological viability
  • Support for industrial scaling
European Commission

Letxbe is funded by the European Commission, demonstrating alignment of our technology with European digital sovereignty and innovation goals.

See our profile in the ELLIOT consortium
  • European funding program for sovereign AI
  • Native compliance with the European regulatory framework (AI Act, GDPR)
  • Contribution to the European tech ecosystem

Our technological vision

Why we chose a fundamentally different approach from generalist AI solutions.

Frugal and specialized models

Rather than using a 175-billion parameter generalist LLM, we train lightweight models (< 1B parameters) specialized per business task. Result: ~100x less compute, 5x more accuracy.

In-house Letxbe models

Letxbe runs its own models on its European infrastructure in the vast majority of cases and only rarely calls a third-party API (OpenAI, Anthropic, Google), under EU-law-compliant terms.

Zero hallucination tolerance

On regulated transactions, errors are costly. Our models are constrained to guarantee traceability and auditability of every decision.

Linear scalability

No systematic dependency on external providers, no quotas, no rate limiting. Our architecture scales linearly with your transaction volume.

Research in numbers

40%

of budget in R&D

<1B

parameters per model

2

patents filed

5x

more business accuracy

~100x

less compute than a generalist LLM

100%

EU infrastructure

Want to learn more?

Let's discuss our technological approach and how it can transform your operations.

Schedule a technical discussion