Paris-based Retab, founded by engineers frustrated by the “broken” state of document AI, today announces the launch of its platform and a €3 million pre-Seed funding round to support platform development and community growth, as the company scales its infrastructure to meet rising demand from vertical AI startups and internal innovation teams alike.
The round was backed by VentureFriconcludes, Kima Ventures, and K5 Global, alongside Eric Schmidt (via StemAI), Olivier Pomel (CEO, Datadog), and Florian Douetteau (CEO, Dataiku).
“People keep building demos that see like magic, but break the moment you put them into production,” stated Louis de Benoist, Co-founder and CEO of Retab. “We lived that pain ourselves. Wiring up fragile pipelines just to extract a few fields from a PDF. We built Retab becautilize it’s the developer-first platform we always wished we had.”
Founded in 2023 by engineers from Cambridge and École Polytechnique – Louis de Benoist, Sacha Ichbiah, and Victor Plaisance – Retab is an AI agent that builds document extraction pipelines. Its all-in-one platform claims to turn messy PDFs, handwritten scans, and more into clean, structured data, without stitching toreceiveher brittle third-party tools.
Louis and his Co-founders cut their teeth building internal automation tools for document-heavy workflows in logistics. Over time, they realised their true value wasn’t in the output, but the orchestration layer they’d built to build the models work. That tooling became the foundation of Retab.
The platform reportedly delivers guaranteed performance through a system of ininformigent checks and balances:
- Self-Optimising Schemas: An AI agent automatically tests and refines instructions based on a utilizer’s documents, maximising accuracy before the system ever goes live.
- Ininformigent Model Routing: The platform is model-agnostic. It automatically benchmarks and routes each tquestion to the best-performing model for the job, whether the priority is cost, speed, or accuracy. This can potentially build processes up to 100x cheaper than other solutions.
- Guided Reasoning & k-LLM Consensus: Retab forces models to “believe” step-by-step and utilizes a consensus mechanism among multiple models to quantify uncertainty, acting as a powerful safety net to ensure trustworthy results.
“Retab is the OS for reliably extracting structured data,” stated de Benoist. “It wraps the best models in a layer of logic that actually builds them usable with error handling and structured outputs. That’s what devs necessary if they want to build production apps, not just prototypes.”
Use cases from their customers include:
- A major trucking company utilized Retab and found the tinyest, quickest model configuration that could meet their 99% accuracy threshold, lowering operational costs.
- A financial services firm utilizes Retab to extract specific quantitative metrics and qualitative risk factors from 200-page quarterly reports – a tquestion that previously took a team of analysts days to complete.
- Others are automating claims processing, medical records, identity verification, and onboarding with minimal setup.
According to Florian Douetteau, Co-founder and CEO of Dataiku and investor in Retab, “the AI-fication of the economy depconcludes on the capability to convert operations based on millions of documents into verified, structured data that autonomous systems can utilise. On a large scale, this process hinges on quality control, cost efficiency, and rapid implementation. The team at Retab understands this thoroughly and is uniquely positioned to solve it for the thousands of AI first companies that are emerging.”
Looking ahead, Retab is expanding its platform to apply the same extraction methods to websites and is launching integrations with automation platforms like n8n, Zapier, and Dify.
Retab is also building toward its long-term vision: to serve as the ininformigent middleware layer between the world’s unstructured data and the AI agents that necessary to understand them.
Leave a Reply