Kairntech was born in January 2019
to guide Machine Learning practitioners on path to production.

Posted by Vincent Nibart on August 26, 2019

Kairntech was founded in January 2019 by a team of engineers specializing in artificial intelligence and software development. Providing a Machine Learning Engineering Platform based on cutting edge open source components and dedicated to unstructured content, kairntech supports key accounts, administrations, content providers and startups in the implementation and deployment of Artificial Intelligence products or services. This platform allows our customers to accelerate their digital transformation by empowering their employees and streamlining business processes.

Why the name kairntech?

Combination of "kairn" (cairn but with a k for "Knowledge") meaning the pile of stone guiding you along a mountain path and "tech" - technology - which is our DNA. Kairntech is here to guide you on the path of implementing and deploying Machine Learning based solution applied to unstructured content!

Why Tarmac / Inovallée at Grenoble?

Because the majority of the team is from Grenoble, France. This is a great place to live, a modern, thriving technology location, a dedicated French hot spot in Artificial Intelligence in the midst of impressive mountain ranges. We wanted to benefit from the support and assistance of an incubator. Tarmac is one of the digital & high tech incubator of Grenoble, part of the "French Tech in the Alps".

Why kairntech?

To meet the growing need to implement Machine Learning & Deep Learning products or services on unstructured content (text, audio...) for the Enterprise. We want to help our customers creating data-driven products to generate new revenue streams or to be more effective while leveraging the latest progress in Machine Learning & Deep Learning technology applied to Natural Language Processing. Indeed, it seems 2018 was a pivotal year for Machine Learning technology in NLP. It appears that quality and effectiveness of these approaches have - today - surpassed traditional rule-based methods (but don't get me wrong, we have extensive experience in linguistic NLP, too, such we can combine approaches where appropriate). Creating and maintaining rules used to be a challenge if not a blocking point... today when using "Machine Learning"-driven approaches, the burden has shifted to the creation of high quality learning data sets, required to take a project from implementation to production quality and deployment, delivering products or services quickly...

Finding the best learning model on an existing training dataset is one thing.... implementing and deploying ML-based solutions from A to Z is another thing! Indeed, it's a question of addressing the following steps:

  • Define first an AI product or service,
  • Formalize the learning problem,
  • Collect and explore relevant and representative unstructured content,
  • Define annotation classes & guidelines,
  • Initiate the learning data set thanks to a text annotation tool, version
  • Build a first simple model, test, optimize, measure, experiment, version,
  • Extend the learning data set, improve the first model, test, optimize, measure, experiment, version,
  • Iterate... and lastly validate the model,
  • Test the production pipeline with the developed model(s) and other possible components, test performance,
  • Deploy pipeline in production (customer workflow...),
  • Maintain & improve the solution over time (learning data set and training models) by developing feedback loops...

This is why we have decided to develop a Machine Learning Engineering platform to equip and support our customers in creating new ML-based products or services leveraging unstructured content.

Indeed, we believe that providing human experts with automated AI-driven support is what needed in many scenarios today. Human experts who get appropriate machine support are, for many tasks, the most efficient solution in terms of productivity, efficiency, and quality.