Text mining, semantic extraction

a powerful foundation for all your text analysis needs

Thanks to the unique ability to learn high quality representations on large amounts of text (unsupervised or semi-supervised, if you have class data, for instance), without sacrificing the size of the vocabulary, eXenGine can be used to perform a large variety of tasks : classification, clustering, relevant terms extraction, similar documents search,...

Read more

Relations and behaviour analysis

to connect the dots

Large scale connections analysis is a hard task, and usually Matrix Factorization methods are quickly discarded because they are too expensive, or have a low quality. Other methods like graph-based distances are more scalable but miss a lot of generalizations and are more prone to hubness problems. Our very special algorithm, NCISC, is one of the very rare that combine speed, scalability, robustness and quality

Read more


remove all limits

Text or behaviour analysis of high quality used to be expensive or just undoable : methods like Deep Learning (word2vec or GloVe) and LDA are very hard to scale or too expensive. With our data analysis engine eXenGine and our Wikipedia exploration demo, we have proved that performing an analysis of the same quality can be done on a full vocabulary, on the full link graph for less than $25

Read more


Web crawling with a semantic focus

Combined with large scale semantic analysis, we have developped a scalable crawling technology in order to power our search engine and provide our customers with already available data. The crawler is part of an ongoing R&D effort with which involves several partnerships on a few vertical segments.

Read more

Webscale search engine

Discover public information in a whole new way

Based on our text and graph analysis technology, and on the crawler being currently actively develop, we are developing a full web-scale search engine focused on data discovery : instead of a pure query-based approach, we hope to leverage the semantic similarity between objects in the WWW to help people discover new territories and get to the gems contained on the internet

Read more


The power of cross-data analysis!

With eXenGine, you bring the unique power of a document analysis solution that unifies text (semantic) processing, graph analysis (inter document links, social connections), and behavior analysis.

Cross-data engine

eXenGine uses a unique machine learning algorithm with extreme robustness, high relevance and the unique ability to cross-feed data from different sources. eXenGine can be used as a foundation for recommender systems (behavior or content-based), document exploration, classification, customer segmentation,...

Big Data

Based on Apache Spark, and more importantly, using a super-fast, super-scalable analysis method, eXenGine can deal with several hundreds of millions of documents, and provides a real-time query engine for the exploitation of the analysis. There is simply no limit to the amount of data you can process

Your use case

We provide extreme customization services to adapt eXenGine to your problem and help you beat the competition in your area : search, e-marketing, law, patents, scientific edition, e-commerce.

News at eXenSa

  • eXenSa at AI Paris – showcasing eXenGine 2

    On June, 6th and 7th, we'll be at AI Paris to showcase our next version of eXenGine, combining exclusive technologies to push data mining to the next level : Super efficient counting (10 times less memory than current state-of-the art competition) Ultra-fast Machine Learning (we can create a model synthetizing half a billion documents in 30 minutes on a cluster of Continue Reading
  • eXenSa at iSwag

    Guillaume Pitel gave a talk at iSwag 2016 about our newest breakthrough algorithm for fast approximate counting. Our Count-Min-Tree sketch structure outperforms the state of the art methods by serveral orders of magnitude. See the presentation and the paper.  
  • Meet you at the Big Data Business Convention

    University Paris-Saclay organizes the first Business / Academics event in France with the Big Data Business Convention 24-25 October at the HEC campus. Meet you at the eXenSa's booth to talk about our latest works and our future : we're working on a online version of our multi data anaylsis solution. Currently we process the whole English Wikipedia from scratch Continue Reading
Other articles

Our Services

Customizing and Licensing our technology is our primary activity. 
We provide advices, development and report to help you build your own data analysis solutions : Big Data algorithms, Data Science, Machine Learning and Big Data systems (Apache Spark) are our domain of expertise. 
We help companies doing research to write their annual activity reports for tax credits
We give training session on GPGPU, Spark, High performance computing. 

Latest News