Jonathan
Grandperrin

There is a document problem hiding inside every software company on earth. It is sitting in the accounts payable inbox. It is stacked in the KYC queue. It is jammed in the expense report pipeline. Someone - often a contractor, sometimes an intern - is typing the same invoice fields into a spreadsheet that a computer could read in 400 milliseconds.

Jonathan Grandperrin noticed this. He built a company to fix it.

Mindee is an API-first platform for intelligent document processing - the kind of product that lets a developer drop three lines of code into their application and suddenly their software can read a receipt, parse a passport, extract line items from an invoice, and pipe it all into structured JSON. Since its 2018 founding, it has quietly become the document understanding layer for accounting software, expense management platforms, and enterprise AP automation tools across Europe and North America.

The numbers back the thesis. Mindee hit $3.5M ARR in 2024, up 52% year-over-year, with 70 enterprise customers and a net dollar retention rate of 200-250% after year one. That last figure is the one that keeps investors awake (in a good way) - it means customers that have been on the platform for a year are spending two to two-and-a-half times what they originally contracted. Zero logo churn. Negative overall churn. These are the metrics of a product that embeds itself and then expands.

The path to building document AI ran through some unexpected terrain. Grandperrin graduated from École Polytechnique - France's elite engineering school, the one that produces a disproportionate share of the country's CEOs, generals, and mathematicians. He followed it with a master's degree in Entrepreneurship and Innovation at UC Berkeley, which planted him squarely in the Northern California startup ecosystem long before Mindee existed.

Before founding Mindee, he spent two stints as a CTO at early-stage French startups. From 2014 to 2016, he ran technology at ECTOR, a valet parking service. Then from 2016 to 2018, he was CTO at TankYou, a digital platform for fuel delivery and vehicle maintenance. Both experiences put him in the trenches of building web and mobile applications, designing API gateways, and hiring developers. Neither was a glamorous run-up to a billion-dollar vision. Both were exactly the kind of operational experience that makes a CEO better at building product teams later.

Earlier still, he spent six months at PSA Peugeot Citroen working on climate optimization before the startup world claimed him entirely. The arc from French automotive giant to parking app CTO to document AI founder is not a straight line. It is exactly the kind of career that produces founders who know how enterprises actually work.

Grandperrin co-founded Mindee in early 2018 alongside Olivier Rey (CTO), Georges Ryssen, and Mohamed Biaz (Chief of Science). The founding premise was specific: software developers needed a better way to handle document data extraction, and the existing tools - mostly legacy OCR vendors - were not built for API-first products.

Mindee spent its first years in stealth, bootstrapping toward product-market fit. A $2.2M seed round in 2019 provided the runway to refine the model. Then came Y Combinator. The W21 batch - the pandemic cohort - put Mindee in Mountain View alongside some of the sharpest developer-tool builders in the world. The YC stamp mattered less for the check than for the network and the discipline it instilled around metrics.

In October 2021, Mindee came out of stealth with a $14M Series A led by GGV Capital, with participation from Alven, Serena Capital, and Bpifrance Digital Venture. The round included two notable angel investors: Nicolas Dessaigne, co-founder and former CEO of Algolia, and Alexis Le-Quoc, co-founder and CTO of Datadog. That combination - a search infrastructure legend and a monitoring infrastructure legend - said something about where Grandperrin was positioning Mindee in the developer infrastructure stack.

In March 2023, a further $7M venture round extended the runway. Total funding now exceeds $23M.

Mindee's design principle is not complicated. Grandperrin has consistently argued that document AI's fundamental problem was not accuracy - it was developer experience. Legacy OCR tools gave you coordinates and bounding boxes. They gave you a PDF and a headache. Mindee gives you a JSON object with named fields, confidence scores, and a model you can fine-tune on your own documents.

The platform operates on three pillars: a pre-built API catalog for common documents (invoices, receipts, passports, driver's licenses), a custom model training interface for proprietary document types, and an enterprise SDK for teams that need on-premise or deeply embedded solutions. The company also maintains docTR, an open-source document text recognition library that has become one of the most widely used OCR toolkits on GitHub - a developer relations play that puts Mindee's name in the hands of anyone building document workflows.

Grandperrin has framed the competitive landscape with precision. He draws a sharp line between traditional OCR - pixel-level text recognition - and Intelligent Document Processing (IDP), which involves understanding document semantics: knowing that the number on line 7 of the invoice is the tax amount, not the PO number, even when formats vary across vendors.

Mindee runs a transatlantic operation that mirrors Grandperrin's own biography. The engineering and product teams are in Paris, based at 14 Rue Charles V in the Marais. The commercial presence - go-to-market, partnerships, investor relations - runs through San Francisco. Grandperrin himself is based in California.

This dual structure is increasingly common among European technical founders with YC pedigree. The French startup ecosystem, particularly in Paris, has produced serious engineering talent in AI and machine learning. The US market provides the scale and enterprise contract sizes that justify the engineering investment. Mindee's customer base is explicitly transatlantic: think European SMB accounting platforms, American AP automation tools, and global KYC compliance systems.

The company's partnerships reflect this positioning. A tie-up with PayFit, the French HR software company, demonstrates traction in the European expense management space. Integration with the enterprise software ecosystem - Salesforce, Microsoft Azure, Slack, Google Workspace - shows the platform's connectivity ambitions.

Grandperrin is unusually hands-on as a publisher. His personal Medium profile self-describes him as a "Computer vision & software dev enthusiast" - not a CEO persona, a practitioner persona. His early articles walked through VIN extraction with custom OCR models, confidence scoring in ML systems, and Python receipt parsing tutorials. This is not marketing copy by a content team. These are technical explainers written by someone who still thinks in model architecture.

He has contributed articles to VentureBeat, DevPro Journal, and Dataversity, consistently writing about the practical implications of document AI for enterprise digital transformation rather than the hype around AI generally. The posture is consistent: document processing is infrastructure. Build it right. Ship it to developers. Let the enterprise workflows follow.

His public podcast appearances - including the PodSaas interview and the Data Defenders Forum - reveal a founder who talks about his product with the fluency of someone who built the first version themselves. Which, presumably, he did.