Founded in spring 2017 reciTAL is among the first international teams to master and exploit large pre-trained language models.

The namereciTAL” is a direct reference to TAL, or Automatic Language Processing.

Back in 2017, we used Transformer-based language models to offer document processing tools. Thanks to their pre-training on large volumes of data, these models have a language comprehension capability that facilitates document processing tasks. As well as “seeing” the document, the machine can also read it and understand the meaning of words and phrases. These technologies drastically reduce the volume of data required to learn the desired task.

Whether extractive or generative, these models have ushered in a new era for IDP (Intelligent Document Processing).

Today, reciTAL offers a complete LAD / RAD solution for harnessing the power of large language models, on Premises or SaaS.

Industrial vision


The latest generation of multimodal AI (text + layout, integrating pre-chained language models) is a recent but mature technology, which extends the scope of LAD-RAD, notably by opening the way to the processing of sophisticated documents (unstructured, long documents, tables, financial documents, etc.).


In this new era, industrial success relies as much on AI know-how (training, hosting, deploying in production) as on software know-how (offering an application that meets the diversity and requirements of business processes).

Advantages of LLMs (Large Language Models)


LLMs simultaneously extract and standardize information. For example, it is possible to identify the lines of an accounting balance sheet while reconciling them with a business reference framework (general chart of accounts).


LLMs "generalize", i.e. they are able to process with excellent performance documents whose layout or content have never been seen before. What's more, they are not data-intensive, since the largest reciTAL models are finetuned with less than 1000 annotated examples.


Any type of data can be extracted, without having to specify it. The same technology can be used to extract values, paragrpahs or tables.

The team

The reciTAL team has been built around two principles: excellence and diversity.

reciTAL boasts an international team (+10 languages spoken) of the highest calibre (5 PhDs and engineers from top-ranking engineering schools).

To keep abreast of developments in the field, each member of the reciTAL technical team has one day a week to work on an innovative personal technology project.


Gilles Moyse, PhD


Gilles is an engineer from the UTC and holds a PhD in computer science from Sorbonne University.

Gilles Moyse, PhD

Frédéric Allary

General Manager

Frédéric is a graduate of Sciences-Po and HEC, with a degree in econometrics.

Frédéric Allary

Thank you to our partners who contribute to the success of reciTAL