Inside Net AI: A Conversation with Alexis Duque on Turning Research Into Product

I spoke with Alexis Duque, VP of Engineering at Net AI, about how he joined the company in its earliest stages, what it means to turn research ideas into products, and why AI in telecom needs to be approached carefully. We also discussed PAISES-6G, the limits of full automation, and why sometimes the most useful role for AI is not to replace humans, but to help engineers manage complex networks more effectively.

Life in Lyon and Joining Net AI

Mariam: Tell me a bit about where you’re based and what’s the best thing to do there.

Alexis: I’m based in Lyon, in France, and my favourite thing to do is to get my gravel bike and go to the forest or the Monts d’Or. There are some small mountains nearby, and it’s beautiful. I think what I like about Lyon is that you are in the city, but you can be in the countryside very quickly.

What else can I recommend doing in Lyon? It depends. I like the Fine Arts Museum there. I think it’s the second biggest in France after the Louvre, so it’s a big one. Culturally, there’s a lot to do as well, especially around music. Lyon has a great opera house and a really nice auditorium.

Mariam: That’s great. Cycling seems to be a bit of a Net AI theme, especially between you, Paul and Rokas. Lyon sounds like a good place for it too. Can you tell me a bit about how you joined Net AI? What was your first introduction to the company?

Alexis: Yes. So much time has passed since then. It will be six years in September.

I was thinking about what I wanted to do with my job. I discussed it with my friend, Razvan, who’s been my PhD advisor, and it appeared that he knew Paul, because he’s Romanian as well, and they were both working in the telecoms field.

During my PhD years, I always wanted to work with Paul in some way. He was not yet working in AI and was more focused on IoT and cybersecurity, and I was also doing that in my former company. So I was closely following his work at the University of Edinburgh.

At the time, Paul was looking for someone to start the entrepreneurship process because the company was not created yet. It was just an idea. They had the patent and the paper, and they wanted to make a company based on that. He got a grant from Innovate UK through the ICURe programme, which helped him begin the process of market discovery and setting the company’s direction.

When Razvan found out about this, he thought of me. I was excited about the idea of creating a company from scratch. I was also looking at AI applied to the Internet of Things, because that was my main focus. And I was excited about working on AI for telecoms as well, working with Paul.

So I had an interview, and we started working together in September 2020, during the COVID-19 years. Because everything was remote, I didn’t have to move to Edinburgh, which probably made it possible for me to take the opportunity at that stage.

I wasn’t against the idea of moving to Edinburgh in general, but at the time there wasn’t really a company yet. My contract was only from September to December, and if things went well, it could be extended until around May the following year. So moving country for something that was still so early would have been difficult.

Mariam: Were you already done with your PhD when Net AI started?

Alexis: Yes. I defended my thesis in 2018, so two years before that.

What a VP of Engineering Actually Does

Mariam: What does your role as the VP of Engineering involve?

Alexis: I do a lot of things, and I think VP of Engineering is a bit of a shortcut. My main responsibility is to shape the technical team and the engineering process, and to help the team do their best when writing code. So I spend a lot of time reviewing code.

I give them ideas and unblock them when they’re stuck. Another big part of my role is turning ideas into something industry-ready. I mean something that is engineered, not just a research paper, but software that can actually be packaged and used. So, in a way, I help turn research papers into software.

I also manage a couple of European projects. I attend meetings with partners, prepare deliverables, join technical discussions, and distribute the work across the team. So I absorb a bit of the administrative side of the European projects: the meetings, the coordination, and so on.

Also, when I have time, and it’s a bit of a background task, I look at what the industry is doing, what competitors are doing, and what is happening in standards. That includes what people at O-RAN and 3GPP are working on, how it works, and how we can quickly identify which API or interface we need when a customer has a specific requirement.

Turning Research Into Software

Mariam: You work closely with Orlando, Sam, Ana and Alec in ML research and ML engineering. I had this question further down the line, but since you pointed it out already, Net AI is very closely connected to academia, and a lot of the team’s work starts from research, ideas, and papers.

I’d like to talk about how that translates into the business world. You mentioned turning ideas into a technical product. How do you turn research into something operators can use and explain what we do in simpler terms without oversimplifying it?

Alexis: Yeah, I think we can see research ideas a bit like a raw diamond. To be able to sell it, we first need to polish it. I would say that’s partly the role of the ML researcher and ML engineer team. Then we need to package it and make something nice, something that shines, a good package to send to the customer.

That’s not the whole process, but it’s quite exciting and interesting to refine a research idea and make it something that a customer can buy, while also being able to simplify it and explain it without making it too complex, but also not too simple.

Mariam: And what is the step-by-step for that? Does an ML researcher, for example, read a paper and say, “I found this cool thing, maybe we can integrate it into our solution”?

Alexis: Yes, I would say someone like Orlando, Ana, or Marco will point out a research paper, look at it, and then brainstorm how we can exploit it, modify it, or use it as it is in our project.

Then Sam would run some experiments and implement the model in our pipelines, train it, and write the code necessary to use it in a product.

So that’s the second step. And the third step is to make it something we can deploy and sell as a product. That’s more the role of full-stack engineers: building the backend to run the model, building the interface, and creating a broader API for the user to interact with the machine learning model.

Mariam: This is just a question that came to my mind, but how does IP work in that case? When research builds on existing papers and ideas, where is the line between learning from the field and creating something new?

Alexis: Yes, because most of the time, when I say we read papers, it’s not about taking one paper and turning that directly into a product. We read a lot of papers, and by reading all these papers and all these ideas, they spark other ideas that we want to explore.

Sometimes we read something that’s not related to telecoms, maybe a paper using AI in a totally different field. By keeping up with what they are doing, and because we know the telecom space, we know how to take inspiration from those ideas, modify them, and make them suit our needs in telecom. That’s a bit the ML researcher part of the job.

Mariam: Yeah, that makes sense. I guess in university, that’s also how writing a dissertation works, right? You do a lot of research into other work, but then you create something unique from what you’ve learned.

Alexis: Yes, exactly.

AI in Telecom: Useful Tool, Not Full Automation

Mariam: Let’s move on to a more general topic of AI and telecom. We’ve talked about this subject extensively before: how AI adoption in telecom is still difficult, and there are a lot of moving pieces, for example, difficulty introducing something new into legacy infrastructure, etc. With this in mind, I wanted to ask you: what is something that people sometimes misunderstand about AI? Where do you think it can actually be useful in telecom?

Alexis: I think initially there were too many expectations for AI. At the beginning, people really believed mobile networks could become autonomous, and everybody was expecting a lot. So the deception was huge. And now they want to put it everywhere.

Of course, it doesn’t work like that. You should go step by step. I don’t believe that right now we are ready to have fully autonomous networks using AI, but I think it can help with very specific tasks.

I think it’s more like a tool for telecoms now. For example, root cause analysis is a nice use case that is easy to implement in real life and in the short term. A Network Digital Twin is for experimentation, trying different network configurations without making them totally autonomous, but just making experiments in a controlled environment that replicates reality.

For energy saving, maybe not putting it as a tool to automate the network and optimise the energy gain automatically, but more to fine-tune the network, even with small gains.

I see it a bit like a coding assistant with AI. We are not asking a coding agent to write the whole code in your application, but you need a tool, step by step, reviewing it and making telecoms more efficient. Not total full automation, but helping people to better manage the network, rather than giving everything to AI and expecting the network to work for free, and that’s it.

Mariam: Do you think the problem with full automation is that it can overlook some issues that a human would not overlook?

Alexis: There is this risk, and also the fact that I think networks are not ready to be totally autonomous right now. They don’t have everything needed for that. Some things still need to be manually done, so it would be a mess. And data is not always collected in a consistent way. It’s very complicated to have something that works autonomously without someone looking at it.

There is also a lot of responsibility with operating a mobile network. I remember in France some time ago, the emergency number was down for a couple of hours. It was just a mistake with a cable, a mistake somewhere in a box, and some people were not able to go to the hospital in an emergency because of that mistake. There is an operator responsible for that.

So that’s also part of what makes it frightening to make something totally autonomous when you have such huge responsibility. Quality of service, financial penalties, and image are all involved.

Mariam: Recently Paul made a point on the Sparring Partners podcast that AI is not really needed for everything, and in many cases it can be an overkill. That’s what prompted me to ask you this question, because we do see it everywhere, and we hear a lot of companies use it all the time.

But again, this could be a case of big hype around it. Everyone spent a lot of money, and now they’re trying to jam it into every type of operation possible to make a return on investment.

Alexis: Recently, I had lunch with Razvan, the friend I mentioned before, and he was working on something for a company. It was a type of research project, and they were doing root cause analysis. They wanted to use AI.

But the thing is, it was just so simple that it didn’t make sense to use AI. A simple static sequence model did the job. And when the static sequence model didn’t do the job, it was just because they didn’t have the data to be able to get the root cause. So even AI couldn’t do that. Sometimes people want AI just because it sounds good.

PAISES-6G and the Next Step for AI in Networks

Mariam: Can you talk a bit about the work we’re doing across our current projects? It would be nice to touch on PAISES-6G, for example, because it’s newly launched and we haven’t really written much about it. If you’re involved in that, what are your thoughts? What’s happening?

Alexis: Okay, so PAISES just started. We had the kickoff in June, and we had only one meeting two weeks ago. So it’s super new, but I’m super excited about it. It’s a new project with a lot of partners. It’s one of the biggest projects that I have ever participated in.

It’s about cybersecurity for 6G, and anomaly detection will be our main focus in the project. We are leading a task on that, and most of our efforts will be on a work package where we will develop anomaly detection for autonomous networks and agentic AI for telecoms.

So it will probably be some agent to flag a cyber attack or anomalies. That can be a cyber attack, or network misconfiguration, or maybe another model or another agent willing to do something super awkward to the network, and we spot this.

Mariam: Does it connect to a previous Horizon Europe project, like ORIGAMI for example? Because again, it’s funded by them. ORIGAMI’s main message was to tackle the barriers that we’re going to encounter with 6G. And this focuses on cybersecurity. So I was wondering if they’re connected.

Alexis: It’s in the Horizon Europe programme, so it’s like the next step. I think we can, in some way, think that it builds upon some work that has been done in 5G, proposing more AI-compatible 5G and 6G. And on top of that, building a kind of agentic AI into 6G with a focus on cybersecurity.

And Finally: A Recommendation

Mariam: That’s all the main questions I had, but I wanted to end with a recommendation. What is a recent book, film, podcast, or piece of media you enjoyed?

Alexis: I just read a really nice book that a friend recommended to me. It’s a comic, but a serious one, about Iran. It’s the story of a girl living through the Iranian Revolution in the 80s.

Mariam: Are you talking about Persepolis? I love it. It’s one of my favourite graphic novels.

Alexis: Yes! And I think the story goes beyond Iran. It brings a universal message about life, and Marjie is someone many people can relate to in some way.

I also studied the film in high school, so I had some old memories of it. Seeing the story again almost 20 years later gave me a very different understanding. Now I think it’s an even better one than I did back then.

Mariam: I agree. It’s really well written, and with any well-written book, you can find yourself in it in some way. Thank you for that. I’m very happy that’s your recommendation.

Alexis: So you will make some more publicity for Persepolis.

Mariam: Yes! That’s the biggest takeaway I got from this. Thank you, Alexis.

Monts d’Or, captured by Alexis