Retab Exits Stealth Mode with $3.5M for SOTA AI Document Processing

Retab, a developer-first document AI platform, has raised $3.5 million in pre-seed funding and officially launched its product. The San Francisco-based startup, with roots in Paris, aims to become the foundational infrastructure for vertical AI applications by simplifying how developers extract structured data from unstructured documents.

Founded by engineers Louis de Benoist, Sacha Ichbiah, and Victor Plaisance, Retab was born from firsthand experience with the limitations of existing document processing tools. The company’s new platform automates the complex pipeline of data extraction, allowing developers to define the schema of data they need while Retab handles dataset labeling, prompt engineering, model selection, and performance evaluation.

The funding round includes participation from VentureFriends, Kima Ventures, K5 Global, and notable angel investors such as Eric Schmidt (via StemAI), Olivier Pomel (CEO of Datadog), and Florian Douetteau (CEO of Dataiku). The capital will accelerate platform development and community expansion as Retab scales to support startups and internal AI teams.

Unlike traditional large language models, Retab acts as an orchestration layer across providers like OpenAI, Google, and Anthropic. It delivers consistent and verifiable results through:

• Self-Optimizing Schemas: Automatically refine instructions based on document context.
• Intelligent Model Routing: Selects the optimal model for each task based on performance metrics like cost, speed, and accuracy.
• Guided Reasoning and k-LLM Consensus: Enhances reliability by coordinating outputs across multiple models and validating consistency.

Retab is already in use across industries including logistics, finance, and healthcare. One logistics firm reduced costs by identifying a lightweight model configuration that still met their 99% accuracy threshold. In financial services, Retab automates extraction of key metrics from lengthy quarterly reports, significantly cutting analyst time. Other applications include claims processing, identity verification, and medical record analysis.

According to investor Florian Douetteau, "Retab solves the fundamental challenge of transforming document-heavy operations into structured data pipelines that autonomous systems can reliably use. It’s a critical enabler for AI adoption at scale."

Retab is now expanding its capabilities to include data extraction from websites and rolling out integrations with platforms like Zapier, Dify, and n8n. Long-term, the team envisions Retab as the intelligent middleware connecting AI agents to the vast pool of unstructured data that powers global operations.

With just ten employees, Retab is rapidly becoming an essential tool for developers building real-world AI applications—offering production-grade performance from day one.