Beewant - Multimodal AI-Powered Data Management

Tony [00:00:04] Welcome to Code Together, a podcast for developers by developers, where we discuss technology and trends in industry.  I'm your host Tony Mongkolsmai.

Tony [00:00:17] I've talked to several people who are part of the Intel Liftoff for Startups program this year. And today we're joined by another leader for one of our innovative companies. Ahmed Joudad is CEO and founder of Beewant, which is an AI platform that makes it easy for companies to navigate through large volumes of unstructured data. He has spent most of his career working for a global provider of financial market data and platforms. He's always had a passion for technology, particularly A.I. and machine learning, and believes that the key to unlocking the full potential of A.I. is high quality data that would enable algorithms to learn and improve. Ahmed is based in Paris and is a passionate golfer and soccer player in his free time. So welcome to the podcast Ahmed.

Ahmed [00:00:55] Thanks for having me.

Tony [00:00:57] So what's funny is I read your intro and I'd really like to talk about golf and soccer. But today we're going to talk about Beewant. So can you tell us a little bit about Beewant and what your mission is as a company?

Ahmed [00:01:11] Yeah, sure. I mean, Beewant is an AI platform that helps companies interact with their unstructured data. And by unstructured data, I mean, not only text, but also images and videos. If you consider, like an enterprise that has a lot of data. Nearly 80% of the data is unstructured. And right now there is a lack of tools to navigate through that data and to explore it and to extract value from it. So we intend to change that.

Tony [00:02:00] So when you say unstructured data, a lot of people nowadays are thinking about things like large language models where they say, "I'm going to take a large language model, I'm going to take all of the documentation and text I have and throw it into that large language model and then use it kind of as a search engine" similar to how people are using ChatGPT for the for the Internet. How is it different than something like that? Because they're we're talking about text. But I believe for you, you're talking about more than just text.

Ahmed [00:02:31] Yeah, I mean it. If you see the word. I mean, we don't deal only with texts. We deal also with images. We deal with videos, We deal with audio. So, think about like how humans interact with their surroundings. When you start as a baby, you hear you see stuff that probably you don't understand. And then when you grow up a little bit as a kid, you you start asking questions. And then when you become an adult, you tend to specialize. Having just LLM out there is, in my opinion, not sufficient to have a full AGI system. We need to have like images and videos and being able to interact with with, you know, those those digital assets as well. So our mission at Beewant is to have like a kind of ecosystem where you can interact with, of course, language, but also with images and videos, which is a little bit challenging because most of the things that we see right now is is around LLMs.

Tony [00:04:01] Yeah. And for those who haven't seen it at our oneAPI DevSummit earlier this year, we had basically a video of you and Sebastian from Weaviate talking about how you guys were using Beewant's Beesearch to help kind of help kind of index a variety of videos and audio and text from Paris. Can you describe a little bit about how you came about to kind of capturing all that data from Paris and putting that into your system? And then probably what we'll do is we'll try to link maybe that that demo at the end of the podcast so people can see that in action. But can you tell us a little bit about what that was like?

Ahmed [00:04:48] Yeah, I mean, our solutions are not only about like specific domain, the model is able to just interact with different types of data and different types of domains. If you are talking about culture, I mean, we we are based in Paris, and I was lucky enough that the Paris region was interested about, you know, what we were doing. And during the talk, they told us that they had like huge amounts of images and videos, but they actually didn't have any tool to analyze those images and videos. They want to be able to extract metadata. They want to be able to filter like images with people. For example, they want to have some insights on copyright and a lot of like curation. And this is how it all started. So basically what we did is we just tried to collect as much like data about cultural heritage in the region, and we adapted our mixed-model system to the domain of culture to enable that model to detect like concepts, places, even events. And so try to imagine like a visit to a museum. I mean, having a system that could help you better understand the paintings, better understand this cultures better understand, like how it was, which was made by who and what was the context. I mean, it's like a personal assistant with superpowers in terms of like the culture domain. So this is how it all started. And we are very happy about our collaboration partnership with the the Paris region because, you know, it's a vibrant ecosystem for HPC and AI and they are trying to push initiatives like this within the region. And also the goal, of course, is to make those images and videos available to the to the larger public. So we are very happy to to contribute.

Tony [00:07:41] And in that case, when you're for instance, when you said it's kind of like a personal assistant and you you mentioned if you were going to museum, is the is the way that you would search? Is it text only? Could you actually, like, use an image and say, tell me more about this image? What's the the interface that you guys are expecting people to use?

Ahmed [00:08:01] Yeah. I mean, we have both. You can enter queries and search for a concept or like an object, and you can also like insert an image and search for similarity. Probably it's very relevant for culture, but it's kind of more relevant for retail. For example, let's say you are searching for a white shirt with like a blue color or like you went to a friend's house and to so kind of like a sofa that you would like to have in your own house and you don't know how to describe it. So one way to do it is just take a picture and then, you know, and enter that picture to the system and get similarities and probably have a kind of recommendation system that will guide you to where you can buy it or where you can find like the best the best deal for that, that that object. So, yeah, I mean, it has. It has different uses and it's we are still figuring out, you know, how we can use it, what what kind of domains could be relevant for our solutions.

Tony [00:09:28] So you mentioned that you partnered, you actually partnered with the Paris region itself, which is kind of a what's the right word, a government entity. I guess I don't have to say that a lot in this podcast... But you also obviously are looking at how you can partner with businesses as well. Can you take us a little bit inside of how you would think that businesses could take advantage of your technology?

Ahmed [00:09:57] Yeah, of course. I mean. I think about financial services for example. Last time I visited a big bank in Paris and. I was really surprised by the number of use cases they had in the pipe. And the maturity also of they're like brainstorming and they're not brainstorming around, AI, uh, when I started two years ago, financial services were, you know, didn't pay attention to AI too much. o when I see how AI can bring value to financial services right now, that that could be like a very, very good use case for us. I try to think about like if you want to make up like an investment and you want to be able to compare like ESG reports, for example, for two companies and trying to figure out which one you want to invest in. If you had to do that, I mean, you need to read like hundreds of pages of reports, probably returns that you might not understand. And what our what we aim for is to be able to provide, like our users, an easy way to ask questions and get relevant answers, whatever the data. It could be like documents, it could be reports, it could be procedures. It could be FAQs. And talking about the banking industry, I think about regulation. I mean, I don't know  if you know about MiFID regulation. It's really cumbersome to get through, like all the regulation, the changes. And you need to, like, think about the first version, the second one, the third one, and so on. So if you had like a system that could help you to just navigate through all that kind of like documents and regulation, that that actually could be like very, very beneficial. When when we talk about financial services, but we could also talk about like security, for example, I mean, think about like elderly monitoring, for example. We know that's like having a system that could detect to the camera that like an old person is falling or like hurt themselves or whatever and alerts like the authorities are that the people that is monitoring and to be able to save that person. I think about like traffic surveillance, for example. I mean, there are unlimited use cases that could be applied to a multimodal system like like what we are building right now.

Tony [00:13:39] And so maybe if we could dive a little bit into the technology. You mentioned that it's a multimodal A.I. system, which means that you're combining a lot of different models to kind of bridge the gap and create one unified view of the world, essentially one unified heuristic using all those models. Can you talk a little bit about how you guys are doing that and what types of models you're looking at combining together into one single platform?

Ahmed [00:14:10] Yeah. When we started building the solution, we started with images and videos. And when you look at, like, the deep learning models out there, they are focused on one single task and they actually suffer from a like out of distribution. The issue, like, let's say you're considering like detecting hard hats, for example, in this for I don't know, I know a working platform. If you move the camera to like, let's say supermarkets, it will behave, the model will behave differently. And using foundation models for that purpose is a great way to like, solve that problem. And our solution is like a combination of different models, but we tend to keep like one model for images and videos and another model for language. And basically, if you want to go like into technical details, we use contrastive learning, basically images with their descriptions and this is how we fine tune most of our models.

Tony [00:15:47] Okay. And so for each customer, you'd obviously have to go in and and fine tune your models basically to to meet their needs. I'm assuming that would be kind of the service you're looking at providing.

Ahmed [00:16:00] Yeah, that's right. I mean, I do believe that one model couldn't fulfill all the like this specificities that we find at the enterprise level and adapting those models to a specific domain could provide additional value compared to a model that ismultipurpose. So think about retail, for example. If you are building a model for a luxury brand versus a discounter, for example. How do you decide what's expensive? What's cheap, for example? I mean luxury versus discounting, it's two different levels and we tend to give the client like full control over what they want to see in those models. Let's say, for example, you're selling dresses. Who gets to decide if the dress is beautiful or if the dress is like awful. So this is something that we prefer to like, handle to our clients to decide how they want the model to behave given the set of data they have. And so yeah, basically, I believe that each company should have its own AI. Because when you think about it like a company, it's a set of of people of like procedures and then it's your property, of know how. And you can't have one model that will be convenient for every company out there. So I believe there is a way to specialize those models to comply with the culture and comply with like the vision of the of our clients. And this is the way we think about adaptation specialization of those foundation models.

Tony [00:18:45] Okay. And I know that because you're part of the Intel LiftOff for Startups program, you've worked with our engineering teams there. And I also mentioned that you had a partnership with Weaviate that you talked about at our oneAPI DevSummit. Can you talk a little bit about how the Intel Liftoff for Startups program has helped you? I, I think you met Weaviate through them. Maybe you work with them before?  And potentially how did, how did the Intel Developer Cloud help you in terms of building out your models or understanding the performance of your models, etc.?

Ahmed [00:19:19] Yeah, it's a it's a funny story. I mean, we have been building on top of like foundation models since one year now and. A milestone for us was the hackathon, the oneAPI hackathon where during three days we were able to build like the foundation of our system. It started with like microservices architecture, but also how can we scale our solution to millions or even billions of inferences? The oneAPI team was really great because they have introduced us to some of like the acceleration concepts that we didn't know anything about, like IPEX and, you know, all the oneAPI toolkit. And that was really meaningful for us because the, the model was working. Because we are a data company and the relevance of the search was goods, but when you consider deploying that for like millions of inferences per day or per hour, this is a different game. And the team was really good at connecting the dots and helping us like with the architecture of our system as it is right now. And when you think about vector databases, they also introduced us to another member of the oneAPI program, which is Weaviate? And with Weaviate we solved one big issue that we had is that when you try to embed images, for example, you get like a certain number of of embeddings that you can store in a JSON file or, you know, whatever, but going beyond a certain amount of embeddings. It's not sustainable anymore. So this is where Weaviate comes into play, and we were very happy about, you know, how we can use their libraries, their solution, you know, to store our embeddings and have a kind of like because when you use our solution, you expect like to enter a prompt and get like a results right away. You don't want to wait, Right? And this is exactly what Weaviate are doing. We store our embeddings there and then it makes the the query process and the answer process much more easy and much more user friendly.

Tony [00:22:42] Awesome. And so I'm going to pivot a little bit. You being a CEO and founder, a lot of times when I have CEOs and founders on, especially from the Liftoff program. I'm really interested in how you became a founder, what it's like to be a CEO. I'd like to give the listeners a little bit of insight into that. So if you could talk to us a little bit about how you came to decide to founding a company. I guess maybe a multiple part question. What's it like to be the CEO of a startup company? What are the challenges you face? Talk to us a little bit about that.

Ahmed [00:23:19] Yeah. Then maybe some insight about my background. I started as a consultant at the big four, and I, I pivoted very quickly towards financial services, where I spent nearly 15 years at that global financial data provider to name it, Reuters, and this is where I learned how to manage teams, how to deal with clients, how to deal with sales, but also, you know, how a company works, really. You know, how do you make profit? How do you maintain a relationship with with clients, how to how to sell? Because this is something really like challenging for most of the founders to be able to sell your product at an early stage. And, and then quickly, I just moved to, uh, to, to the fintech world and I, I worked for almost one year at one of the, the fastest growing fintechs in, in Europe. And then just decided afterwards that it was time for me to to build something and to build my own company. And this is where I, you know, I had the idea of founding a startup that is between in between like the data platforms. And I was missing something, which was like A.I., And so I wanted something between data platforms and AI. I had some experience, like almost like 17 years experience in data and platforms, but I didn't know anything about data science, to be honest. I needed to start from scratch and just try to learn how how to build like a model, how to. I knew how to code, but not the data science way. And so I remember that my first model was a toy classifier because my kids were like always fighting for their toys. And I thought it could be like, nice to have a model to just, you know, decide, you know, who with the toys belongs to. And I started like, taking pictures of each toy and at the time I had the idea to buy like a workstation with multiple GPUs. Thinking that it could help. And my first model was like a TensorFlow model. You know, I had to deal with records and all of that kind of stuff. And on top of that, their workstation was like on Windows. So, you know, it didn't help really. But at the end, it was pretty much successful. I had the model that is that was that great at classifying toys. And it did give me like some hope on the future. I mean, I told myself that if I can do that, then. What would happen if I have like a full team of engineers and that could build really nice stuff. And in May 2021, I hired my first engineers. It was like in the middle of the COVID crisis. And we started building like a marketplace for data sets and models. The idea behind that was like to let our clients have like a full overview of the building process. So they just order projects and they see how we collect the data, how we curate it, how we manage to build like the model architecture. And it's given them like the ability to, to do insurance in by themselves, to test the model and let us do it like the iterations. It was pretty much successful as a business model until one day one of my team members just introduced us to HuggingFace. I was pretty much like shocked. It got like thousands of models free of charge. So I thought that. We need to find another business model based on high quality data rather than just focusing on the the models itself. And. I believe that, you know, that the research community and the open source community in terms of AI is really vibrant. And we get something between 4000-5000  research papers per day in AI. And so I thought that we couldn't compete with the research community. We couldn't compete with like the open source community. So one thing that I thought it could be really helpful for our clients is how do we take their unstructured data, curate it annotate it, manage it and ingest it into a model. And so basically what we did was kind of like, I saw you get in with your own data and you get out with like your own AI. And this is how everything started. So we built a data platform and then during that time we were experiments in a lot with foundation models, and this was like the start of everything we were did today.

Tony [00:30:55] Okay, that's interesting too, because kind of in your path... And some of the other founders I've talked to, it's interesting, some of them have identified a need and then built that solution and they're still on that path. And some of them, like yourself, you were on a path and then you looked at it and you said, this is not what the market needs right now, or in your case, somebody else is already doing something similar. And there there may be a little bit ahead and so you shifted. So that's a an interesting lesson for people who are thinking about building a company, running a company, etc.. So that that's actually a very, very interesting story. So you guys are an AI company, obviously and it sounds like, you know, just based on that story, you've gone through a couple different places in terms of being part of the AI world and business. So let's talk a little bit more generally about AI. As you sit and you think about not just where your company is going, but where A.I. is going in general, what are the things that you think are important in the next couple of years for A.I. businesses and the community in general?

Ahmed [00:32:13] I think safety is really important. I think that. Anyone who is trying to build like an AI model needs to care about safety. Early on, when we started building models, we had to bring like a moderation system. And one of the things that was really surprising is that we were working on drugs datasets. And a lot of pictures in there. Were from a tribe that were always smoking weed. And most of the this tribe were black people. And when we deploy the model, every time the model sees like black people, they just, you know, raise like an alert on on drugs and. Actually, this was, like, uh, different the first time I saw a need to have to prevent that and to have, like, AI safeguards in place. And this applies to everything. I mean, it applies to text, it applies to images and to every company that is working. And I need to, to, to think about that very carefully. And I just try to address those issues early on.

Tony [00:34:22] Okay, So yeah, so you're talking a little bit... at Intel we use, I think we used the term Responsible AI, I think, you know, a lot of people say ethical AI. How do we eliminate biases in our models and our workflows to make sure that the models are actually recognizing what we classify and things that we want them to classify using the right inputs rather than inputs that are adjacent kind of to the core issue that they're trying to identify. It's a really big field. It's really taking off. I think that there's a lot of people who are taking it seriously, which is good. It's interesting, I was talking to somebody yesterday. I'm actually at my my son's gymnastics class who's a who's in a totally different field. But he was talking about how I was going to potentially affect and change the way that we do business. And he even was talking about the the concerns around ethical AI and making sure that AI was solving problems that that we want that want it to solve rather than some other problems that we, you know, some other issues that we think we want to solve but but are not actually solving. So there's a lot of a lot of research and a lot of concern around that, I think in the technology field and actually just kind of in the business field as well. Am I going to get the right business outcome from the AI that I'm using or am I just using A.I. to use A.I. without the right business and other outcome? I think tying those two things together is very important.

Ahmed [00:35:52] Yeah. And then. Yeah. Sorry.

Tony [00:35:55] Go Ahead.

Ahmed [00:35:56] Yeah. I'm thinking about, like. A. So probably for the audience to just share some some of the good practices that we we are implementing at Beewant when dealing with prompts, you cannot expect like every possible like question from the users. So one thing that is very interesting and in text, is that we we can now like put a kind of context to every prompt. Let's say you are in a bank and you want your users only to have answers to legal questions, for example. Okay, So this is something that is possible now and you act on the prompts rather than acting on like the model to to prevent users from asking different questions that are out of context. I mean, if you are in the legal departments, you shouldn't get like answers for what you are going to have for dinner tonight, for example, and so this is something that should be considered in the enterprise world. For images. It's a little bit more complex. So for that, what we do is we apply like kind of like moderation filters and then act on the prompts to prevent, you know, the kind of questions that that could highlight some of the biases that we could have in our models and, of course, unknown biases that that we could have. Like, let's say you want to search for an agreement, for example. I mean, if you're searching for a set of pictures, you don't want like the model to decide who is ugly or who is beautiful in your dataset. So that's that's the kind of like AI safety measures that we are implementing for for our clients, for videos. It's much more complex than images and text because you get objects, concepts, places, but you also get scenes and events and this is definitely an area of research. It's really hard to find a way to do that for videos. We try with the same techniques as images, but still, I mean, it's not perfect enough.

Tony [00:39:04] Yeah, with the dynamic nature of things going in and out of scenes over time, I'm sure that makes the kind of the classification very difficult. Although on the other hand, your model did seem to do a good job in the demo that we saw of Paris. Like when you when we were talking about people dancing, right? It's not like it gives us kind of a a confidence factor so we can see, hey, yeah, the people are dancing here. It's not just showing me the whole video. So in some sense, at least you're doing something right along the way. They're being able to classify what's in different parts of the videos.

Ahmed [00:39:31] Yeah. But talking about AI safety, I mean, what do you think? What what could we do differently?

Tony [00:39:40] I think the, to get back to where you started, one of the interesting things that I think about the most, and it's kind of what you said, high quality data, a part of high quality data a lot of times is trying to figure out how do you get unbiased data, right? What is the right training data set to have? And that's such a big challenge because by its nature, a deep learning model requires so much data. And by almost by definition, the reason why it works is because there's so much data that a human can't parse through it. And if a human can't parse through it, then almost by definition, I can't really make sure that the data is unbiased because I haven't looked at all of the input data. So I think that that is probably one of the biggest challenges in terms of how to make that, how we make sure that the the input data to a model training is unbiased. Obviously, on the other side of that, even if you have decently good input data, you still have the challenge of when I pull data out, like you said, am I prompting for the right things? Am I scoping the capability of the prompt and the model together correctly? And that's also one of the interesting things, is that a lot of people, when they think about A.I. and especially with the explosion of the large language models, the ChatGPTs, which seem to be, I'll say, all knowing which isn't fair. But but they seem that way because it's the way the interaction happens. I think that people feel like an AI model can't answer all of their questions. And I, even when I was talking to the guy mentioned yesterday at my son's class, you know, we talked a lot about the fact that you're going to need lots of different A.I. models because even now, even though we have a chat, GPT and we have a GPT-4, which is a fantastic model, it's got to be fine tuned. Like you said, every business needs their own A.I., every problem space kind of needs its own solution. If I look at it as a very good/great heuristical workflow where an AI is basically a heuristic, just like all all of us are heuristics of our experiences and our knowledge and what we were exposed to. Our A.I. models are exposed to a lot more. Whether the data is good or bad, it's that's hard to tell in terms of what goes into it. Same with people, though, right? Everybody, everybody's the process, you know, their experiences. But I would never go to one person, even the the person that I consider an expert in a field and say make the decision for everybody in the world. Right, because that person is still only the combination of their experiences and they don't have the experience as everybody else. That's why we have things like democracies, republics and things like that, and that's obviously around political structure, but also informational structure. I wouldn't want one scientist saying this is the way the world must be because I am the best physicists in the world. I won the Nobel Prize in physics. That's not the case. We have expertise in all different areas. So I think from that point of view, those are kind of the issues that we need to understand with AI. How do we make smarter A.I., safer A.I., but also make sure that we're building the right AIs on the right problem spaces. And that is something that is going to require people like you building companies who are making things flexible enough that we can make sure our models are really solving the problems that we care about solving.

Ahmed [00:43:15] Yeah. Tony besides like adopting foundation models, you know, like specific domains that you think like we should have specific models for every country, for example?

Tony [00:43:39] I don't know if every country... I wouldn't say we shouldn't. I think, again, it really depends on what the model is trying to achieve. Right. And making sure that it fits into what people what society needed to do. So, again, if I go back to the example of an AI model is a very good example of one person being very smart, having a lot of knowledge, then if I'm trying to make decisions within that country, for that country, then yes, it would make sense to do that. But that's not always the case because, for instance, you know, you live in Europe. There's a lot of the EU, right. Has some type of social structure and cultural structure, even though each country also has its own social and cultural structure. So I think that you you may want one per country, but I think it's less about country lines and more about socio... I don't say socioeconomic, but like sociological lines, right? In terms of people who have agreement and want that model to help them make a decision. Right. We don't want a model that's based on one type of ideology to make a decision for a different group of people and a different ideology that I think that's kind of the challenge. And right now, a lot of the models are tuned specifically around probably European and American type of value systems, which may or may not be what other countries want.

Ahmed [00:45:13] Yeah. Speaking of that, I mean. What we have tried to to to achieve during the process of fine tuning foundation models is try to maintain a balance between like the global knowledge that we could find and find in like a foundation model and the specificity of like the domain that we are targeting. And this was the have a very hard task because when you fine tune a model, the next version tends to forget like the global knowledge. And you need to to maintain that that base knowledge without necessarily like sacrificing the performance of the the task that you are trying to achieve for a domain specific. So this this this was one one area of research and developments that we have started early on. And I think that we've, you know, we will just continue on the path of finding new methods to just combine the global Internet knowledge, which is like this, the specificity of each domain, and also like try to find ways to to avoid bias or just to eliminate bias or probably just, you know, to to, to give control to it to to our clients to define, you know, what is what is bias, what is not.

Tony [00:47:10] So we're getting close to the end of our time. So normally I ask people and I'll ask you anyway, you know, kind of where you hope technology's going, where you think your business is going, etc.. You've kind of talked a little bit just there about the research you guys are looking at going forward. Is there something more broadly that you're interested in terms of technological advances in the AI space or in general?

Ahmed [00:47:37] What I think. I think the future of this is bright. I think that in the next few years we will see a lot of edge devices with AI embedded and. I think that AI will keep amplifying like human intelligence and act like a multiplier of productivity in different domains. I am just concerned about two things. The first one being humanoid robots. It seems like science fiction, but. I think if there is any danger around AI that should be like humanity or robots. Because at the end when you think about like AI. It's mainly like serve a server or a PC or, you know, whatever a device that could be switched off or, you know, just just terminated. But when you think about robots, you think that if you put enough intelligence in those robots, that could be like a danger for humanity? I'm really concerned about that. Not not so much concerned about the AI and how it could evolve, but but really concerned about robots. And the second thing. It may seem like paradoxal, but I think that for everyone to have access to AI advancements and being able to build like tools so quickly and so easily. If you have like bad intentions, that that could be a danger for, you know, for humanity or for for like other startups that are trying to build something like new, innovative for the good of humanity. So these are the two dangers that I see. But otherwise, I'm really confident that, you know, I could be like, help us as humans, be more productive, be more like creative, and, you know, just contribute to to to make the world better.

Tony [00:50:19] Now, one interesting thing. I thought you were going slightly differently. Normally I would just close, but you mentioned kind of the the people building AI bad actors. I thought you were actually going to say the accessibility of AI. So when you think about like Africa as a continent, just the availability of technology is less, which means then there's less input there in terms of, like you were saying with the countries and model issue. Right. Less input in terms of how models are used in Africa, represent Africa and the potential for it to create an even larger divergence in terms of how efficient people are and socioeconomic divides. But that's a different conversation, which we don't have time for today.

Ahmed [00:51:04] Yeah, I I'd like to to just add I like a thought about Africa. I really think that's AI should be inclusive and I think that Africa should be part of the AI revolution. Right now I see Africa playing a role in the data space, in the annotation space. But I believe and I'm confident that Africa will play a much bigger role in the future. And, you know, try to come up with like some creative and innovative solutions like the West. And one of the first things that I did as I as a founder is have like a subsidiary in Africa, because I really believe that, you know, it's a continent that's that has a lot to offer and that it shouldn't be like left off part. It should be part of the AI revolution and I'm really willing to to to just, you know, contribute at my scale.

Tony [00:52:22] Oh, that's awesome. Okay, well, I think that ends our time today. Thank you Ahmed for joining us.

Ahmed [00:52:29] Thank you so much for having me.

Tony [00:52:32] And thank you, our listener, for joining us. Join us next time when we talk more technology and trends in industry.