Bondzai: industrial AI that learns from humans

The fruit of cutting-edge university research straight out of the Institut Alexander Grothendieck Institute (UM) and an entrepreneurial approach focused on innovation, the Bondzai start-up has developed an artificial intelligence system capable of learning live and continuously in a complex industrial environment. It is currently preparing to raise an ambitious €2 million. Interview with co-founder and CEO Alain Fanet.

What is the nature of the link between the Bondzai start-up and the University of Montpellier?

It all began with the work of Bijan Mohammadi, professor of mathematics at the University of Montpellier and researcher at the Alexander Grothendieck Institute. In 2016, he asked himself how artificial intelligence could be used to simplify mathematical problems associated with numerical simulation. He used his 30 years of experience in this field to automate and optimize AI learning processes, resulting in a new learning algorithm in 2018, which he informed Satt AxLR about. 

What led you to co-found a company based on this academic research?

I was looking for an AI project that broke away from the dominant path in order to make artificial intelligence more accessible and simpler. At the end of 2018, Satt AxLR told me about Bijan Mohammadi's project, and in early 2019, they proposed that we do a one-year pre-maturation program to assess our ability to work together. We come from two very different worlds: academia and private enterprise. At the end of 2019, we then started a maturation project with a budget of almost one million euros. Then came the health crisis. On December 28, 2021, we finally created the Bondzai company with all 6 people who are still there today as shareholders. Bijan Mohammadi also took part in the creation of this start-up via a scientific competition (see Décollage#3 Vers le concours scientifique). We had an algorithm and first demonstrators centered on voice, i.e. voice control, which targeted the industrial environment. 

The artificial intelligence algorithm you've developed has specicificitWhat are they?

There are three of them. The first is that our algorithm has the ability to adapt to the dynamics of the context in which it is used, i.e. to create AI on the spot and in real time. Being able to learn in a matter of seconds means it can integrate changes, including untimely variations such as noise in the case of a voice command, and thus continue to identify the person speaking and giving a command. The second specificity of our algorithm is that it learns from very little data, and a simple interaction with a person is enough to correct the AI if it makes a mistake. The AI will take this information into account, whether positive or negative, and reinforce its knowledge. Its third specificity is that our AI doesn't need a server, a cloud or infinite resources to make its calculations. It is embedded on the device and therefore autonomous. This means you can put it in your car, on a calculator or on a small electronic card... 

What markets is your technology aimed at? 

First of all, there's the Industry 4.0 market, which concerns quality issues such as anomaly detection, and predictive maintenance. We are working in particular with Renault on these issues. We analyze the noise of operations carried out by the machine tool to determine any anomalies, and we are currently upgrading our solution to include image analysis. The second sector we're targeting concerns applied human-machine interfaces(advanced human machine interface), which replace the keyboard with gestures to control screens, for example.

What products do you plan to market? ?

We're introducing DavinSy, an embedded software system and the Maestro configuration tool. Our aim is to sell licenses and thus make our customers, manufacturers of industrial iOT-type connected objects, autonomous when it comes to AI development.

What challenges does Bondzai face today?

The first challenge is financial: we are currently preparing our first fund-raising round of 3 million euros. To do this, we need to convince the financial world that our technology, which represents a break with the dominant AI model, is a strong vector for progress. Once the fund-raising is complete, we aim to increase the number of employees to nineteen. Our second challenge is to do everything we can to make the product marketable by the end of 2023-beginning of 2024, so as to rapidly generate recurring sales. If all goes well, this diversity will lead us to expand abroad and into new industrial sectors as early as 2026. Our software solution is a fundamental breakthrough invention, at the heart of many devices, which means that our market is global.

The issue of AI is the source of many fears, particularly since the tidal wave unleashed by the ChatGPT phenomenon. How do you respond to this?

The question of AI awakens fears, which leads to total incomprehension. The aim of our AI is to make things of better quality, while making the most of human knowledge and know-how. In quality control, our AI dialogues with operators, who can correct and improve it by telling it when it's wrong. Our aim is to put people at the heart of the learning process, so that they can pass on knowledge, and not just to the machine.