Sign In

Luc Legardeur

Co-founder. Our Data Catalog is used by 10.000+ data citizens across the world. We help our customers build data democracies where our competitors deploy data monarchies, aristocracies, dictatorships, or anarchies.

Luc Legardeur is a prominent figure in the data management and analytics field, currently serving as the CEO and Co-founder of Zeenea, a company dedicated to enhancing data governance and discovery for organizations. He has been with Zeenea since January 2017, where he focuses on helping companies leverage their data effectively to drive business insights and decisions.1

In addition to his role at Zeenea, Legardeur is actively involved in various other ventures:

  • Shareholder and Board Member at hymaïa since February 2022.
  • Investor and Board Member at dotsha since April 2021.
  • Co-founder of several companies including Texeï, a Salesforce consulting firm, and WeScale, which specializes in cloud platforms.1

Legardeur has a rich history in technology and consulting, having held significant positions such as:

  • Vice President at Publicis.Sapient from November 2018 to January 2020.
  • Owner and President of Xebia France, where he led the company until its merger with Publicis Sapient in 2018.1

His earlier career includes roles at major companies like Oracle and Microsoft, where he gained extensive experience in sales and technical support.1 Luc Legardeur is also recognized for his contributions to the Canadian Data Community, indicating his active engagement in industry discussions and networking.2

Highlights

Mar 13 · linkedin

The Aladdin of AI.

It's fun to watch the AI craze as a spectator.

Lately, AI has been served in every possible sauce.

We are witnessing many announcements, new (often hazy) concepts emerging, and demos of features that, in real life, don't work, serve little purpose, or whose value creation depends on too many prerequisites pushed by the vendors you already know.

At Zeenea, our Ph. D.s have been working with a French research laboratory for over two years to explore different models and approaches.

We experiment extensively and discard a lot. We focus on AI's value-creation aspects, not on generating buzz.

Here's a small inventory of our generative AI findings:

Generating Business Definitions

The automatic generation of a Business Glossary is very simple (a basic task for LLMs) and completely unnecessary. Unless trained on your specific corpus of documents, a generic LLM will produce general definitions (ask ChatGPT for a definition of the word "Customer," and you will understand the problem).

Generating new dataset descriptions

An attractive idea, but far trickier than it sounds. If the model can only rely on structural metadata (put simply, the name of the table and its columns), the descriptions it generates will be tautological more often than not ("the email column of the customer table contains the customer's email, in the form of a character string"). It's very wordy and quite pointless. The added value of a description is precisely to provide information that is absent from the structure and yet indispensable for exploiting the data. Of course, you can increase the model's creativity, but the result will simply shift from tautological to fanciful...

We have, therefore, decided to discard this approach. We continue working on more sophisticated models fed by a richer context - the results are promising, but we consider their reliability still too fragile for widespread use.

Chatbot

Like many others, we're also experimenting with a Chatbot - with the promise of being able to query the catalog in natural language. Our first tests made us laugh. The problem? Scaling up isn't that simple... our chatbot remains sensitive to the quality of the catalog's modeling and to the quality of its semantic layer. These are difficult prerequisites to meet, particularly in the early stages of deployment. Indeed, these early stages are critical for adoption: an erratic chatbot will compromise the trust of new users. We've got some ideas for solving this equation, so stay tuned!

Despite GenAI's enthusiasm over the past two years, we are making modest use of it in our product. The content generated in Zeenea currently sticks to summaries of lengthy descriptions and definitions—this allows us to display them in contexts where the full text is too long (tooltips, search results, previews, etc.).

Winfried Adalbert Etzel, Karl Ivo 🎧 Sokolov , Gregor Zeiler, Ole Olesen-Bagneux, any thoughts?

#AI #datacatalog

Related Questions

What inspired Luc Legardeur to co-found Zeenea?
How has Luc Legardeur's career evolved over the years?
What are some key achievements of Luc Legardeur in the tech industry?
How does Zeenea's data governance approach differ from others?
What are the main projects Luc Legardeur has worked on at Zeenea?
Luc Legardeur
Luc Legardeur, photo 1
Luc Legardeur, photo 2
Add to my network

Location

Greater Paris Metropolitan Region