#424 From Algorithms to AGI: The AI Evolution with Rabeeh Abla

[00:00:00]

Mehmet: Hello and welcome back to a new episode of the CTO Show with Mehmet. Today, I have a guest, which is a little bit kind of a long story. So I have with me Rabeeh Abla, and fun fact, Rabeeh used to help me when I was at the [00:01:00] university. Uh, he was one of the people who helped me actually understand, uh, the object oriented languages, C at that time.

Mehmet: Rabeeh, it's very nice to see you with me on the CTO Show. And you are an entrepreneur, you are also a, a nerd in ai. But I'll leave it to you to just introduce yourself, tell us a bit more about you and also about CSP, and then we can take it from there.

Rabeeh: Okay? Thank you, Mohammed, for this, uh, for hosting me.

Rabeeh: Uh, I, I remember in I think 20, uh, 2000. Two from 22 years. I think. Yes. Yeah. First time we met, uh, in the, uh, C plus plus, uh, I was the lab instructor, uh, there and, uh, helping, uh, you are studying CCE. I remember computer engineering back then. And I started as a [00:02:00] computer scientist doing, uh, Like, uh, lab instructor doing, uh, working as an A.

Rabeeh: I. Research assistant with, uh, Dr Abbas back then. And, uh, I was in love with, uh, with the technology and, uh, A. I. As many other, uh, people in the field. And, uh, I started my journey as a computer scientist and, uh, research assistant. I was a research assistant in the computer aided design and, uh, as well in AI.

Rabeeh: So we worked on many interesting things, uh, back then. Related to a three D surface interpolation. We were working on the cat milk Clark. Anyway, algorithm that was used in the Toy Story movie to create the most smooth surface for [00:03:00] ships, cars and etcetera. And this is where we use some, uh, intelligence back then.

Rabeeh: And, uh, my journey went from university to master's in AI and focusing on bioinformatics and neural networks and all these aspects back then. And I, I work on. Uh, for like six years as a solution architect for Lebanese American Company, uh, before I started, uh, CSP in 2009, uh, which is as well, uh, focused, uh, on ai, the, the name as, uh, technically like constraints, satisfaction, programming solution, focusing on, uh, computational intelligence, uh, and, and, uh, ai.

Rabeeh: Yeah.

Mehmet: Well, it's, it's nice and pleasure to have you Rabeeh here again. And you brought a lot of memories now just by [00:04:00] mentioning, you know, some of the stuff over there. So you mentioned a lot of fields that as computer scientists, you were attracted to. And of course we cannot be fascinated, I would say by these technologies, but you decided to focus more on the AI part.

Mehmet: So What is special for you at least about, you know, AI that you decided to specialize in and also like, and we're going to talk also about what you're currently doing. So why AI Rabeeh?

Rabeeh: Okay. Uh, I think, uh, it's, uh, as a scientist, uh, For me, and I think for many who I used to work with through the lab, we were, I was fascinated as a science on how the mind work.

Rabeeh: So before I went to computer science, I was thinking maybe to do a psychiatrist or study the mind. The study of the mind is [00:05:00] what attracted me. And, uh, Uh, and how you can create intelligent algorithms. And this is where I focused. I remember even as a bachelors, I used to work for free as an A. I research assistant because the university doesn't allow a bachelor to work as an A.

Rabeeh: I research assistant. So I used to work for two years for free just to learn and it wasn't For me, it wasn't for free. It was like you're learning. So you cannot put a price on learning. And, um, this fascination on I remember my first, uh, my BS project. I wanted to create like, uh, Amazon Alexa. That's what I was from 22 years and more.

Rabeeh: So, um, Uh, I wanted to do that, but my professor convinced me like, no, we have to write an AI interpreter [00:06:00] for an AI language, uh, which is called CSP, which solves computation intelligence problems like timetabling, scheduling, uh, resources and, uh, uh, flights or, uh, et cetera. So with this, uh, love for AI, uh. I started taking my electives in A.

Rabeeh: I. So I took like advanced neural network as a as an elective course with master students in in B. S. in the bachelor. And I took back then the highest grade, and it was a course that master students used to escape. But for me, it was like I was thirsty for A. I. And to learn and. I remember one thing I used to whenever we take a course, let's say, uh, object oriented or AI.

Rabeeh: I used to get the books from MIT and Stanford and [00:07:00] and study what MIT and Stanford is studying. So I used to finish the curriculum that we have in the in the university and then study the two others. And I used to practice how to solve the problem in the fastest algorithm in the fastest time. So I was maybe some student didn't like me because I remember some classes I entered to somebody runs to.

Rabeeh: To the drop and add like Rabeeh Abla has entered the class. Let's go drop the course. He's gonna change the and it was my game for me. It was entertaining. Like I couldn't believe somebody will pay you money to program. It's like, Oh, I know how to program more than I can talk. So it was like, uh, this kind of, uh, maybe I was affected with, uh, 80s Bruce Lee movie, like the Bruce Lee of coding, repeat the code [00:08:00] until you, until you master it.

Rabeeh: Uh, this kind of fascination, uh, which is still in me. Yeah, it's obvious.

Mehmet: I can see, you know, in, in your eyes, you know, the way you talk about it and, you know, of course your, your body language, it shows that Rabeeh, but I would say, you know, and maybe this is a bit of your passion. Plus it's kind of how things moved is AI, like who would be able to believe, because I remember when I was, I took the AI course when I was also, uh, uh, First of all, Rabeeh was studying also as well, by the way.

Mehmet: So, uh, I get fascinated, but you know what? Even some of the professors, they were like kind of, yes, theory more than practical, right? So, and all of a sudden, I'm not all of a sudden, of course it took, as you said, 20 plus years and AI is there since the 1950s when they did the first, uh, [00:09:00] you know, gathering, uh, in Dartmouth, right?

Mehmet: So, but for you, I think. What happened later is now everyone is talking AI. All of a sudden, you are the most wanted man, Rabeeh, I believe, because your knowledge, really, if we leave the jokes aside, so maybe my generation, um, and I am, I don't like to hide ages. So we were like, They teach us, you know, something which is the basic of AI, but it's advanced.

Mehmet: Don't get me wrong. So they teach you, for example, how to design a tic tac toe game and the algorithm behind it, also the chess. And we used to use the Lisp programming language. But here we go today, Rabeeh, and we are talking about LLMs and, you know, all these advanced stuff. So for you, this also shifts within the AI itself.

Mehmet: You know, have you been surprised by [00:10:00] like you said, Oh, my God, like this is even faster than what I thought or because you were too much and you were telling me you were studying the books from Stanford and M. I. T. Or no, you said, Okay, this is the natural progression, actually, of how I should become. So how this between then and now, if we want to put it forward.

Mehmet: Yeah. From your point of view.

Rabeeh: Okay, that's a good question. So being close to the A. I. Market. Um, I worked in my business and speech recognition, medical speech, facial recognition. We worked in the cognitive. Uh, I think what moved very rapidly and evolved. Uh, I can put it into two categories. First, uh, data science part.

Rabeeh: So I remember from 20 years, we used to call it like a neural network. And there were the algorithms were there. Now that the science is a [00:11:00] fully structured data science. Major, which is that the scientists can work and and study. It's a combination of statistical algorithm within your network. So I look at the data scientists working in the company.

Rabeeh: They are working at the higher layer. So when they solve the problem, they already know that There are, let's say, 50 algorithms to solve it. And their approach is more of an experimentation versus us. From 20 years, we maybe did a PhD paper on like a neural network, for example, and which, uh, how many layers it should have.

Rabeeh: And we were constructing it from scratch and writing it, whether in MATLAB or C or C sharp or Java. So I remember I wrote a paper. My own neural network libraries. Now that the scientists have machine learning operation platforms, you call them the ML ops, they can benchmark or auto ML. They call it, you give it the data set and it [00:12:00] writes the, it selects the neural networks and it benchmarks them.

Rabeeh: So what. Took us five months to do is being done in one week, so it's all about the data and that the science got established a lot and it's going now like we are using the auto ML and we're using the experimentation. So there is now a full field that is very mature for the last, let's say. At least 10 years.

Rabeeh: And the other part which was missing in the AI was the LLM. So the involvement of the LLM gave the market a quick win to witness how AI can become a useful tool. So that was the surprise for all of us because the previous approach and we're talking about Knowledge basis and knowledge graph, which were there since maybe more than 50 years, they used to take that the [00:13:00] scientist and that the engineer, lots of time to build the graph and to understand the domain.

Rabeeh: Now you don't need to understand the domain. You can just, uh, train that LLM and it will create that links internally. And this changed the whole process. So the breakthrough that happened in LLM. Transformed for us the inference and the summarization. So if you look at GPT, it's doing many things. So it's it has an LP natural language processing and understanding, and it is doing some kind of inference and it's doing even sometimes certain operations.

Rabeeh: So behind that. The big box. There is more than an LLM and it's becoming more of a useful tool. That's the missing gap that surprised us. And when that LLM was in the market, the first thing I did in [00:14:00] CSP, I created Two AI startups from two years started investing in them because for us, that was the last piece missing for us.

Rabeeh: Because if you look at speech recognition, uh, for English, for example, it was hitting accuracy, 98 percent for, for the last 10 years. And then you want, uh, the company, uh, new ones, uh, bought all the engine technologies in 2008 from Philips. So they bought something called the speech magic. And my company was selling speech magic in 2013.

Rabeeh: So Nuance bought all the top speech engines and they were performing accurately since more than 10 years. So there's nothing that, uh, The new and speech recognition, uh, there's something in you and the text to speech. So text to speech became more human like. So I can copy your voice [00:15:00] note, uh, with the 10 second voice.

Rabeeh: And this is as well changing the speech recognition industry. And I'll tell you how. So if I wanted to create a speech engine, for example, and, um. Uh, in English and to train it to for lawyers. I used to have to hire lawyers, take them to a studio and have them, uh, talk maybe hours of text. Now I can clone your voice and I can generate the hours of text.

Rabeeh: So the whole game has changed. So if I wanted to create a speech engine, I used to maybe invest, uh, uh, maybe more money. A lot of money to to make it accurate. Now I can generate realistic synthetic data and I can train that engine. So there were missing gaps in the market, like the voice cloning gap and the LLM these.

Rabeeh: Last two missing pieces, uh, catalyzed the [00:16:00] previous A. I. So they made like if you look at, uh, for example, Chad GPT, the mobile app, you're talking to it. That speech recognition is the same rhythm from 10 years. Nothing changed the text to speech. The human voice. This is four years old. The voice cloning, the realistic, uh, the NLP is more than 15 years old.

Rabeeh: The LLM, let's say, the one they have is like four years breakthrough. So it's, uh, you are assembling an ai, uh, to give value and it's all coming together like to create a human AI human. That's that stage. Uh, and this is I think where even, uh, open ai, uh, two or five years plan, uh, the five years plan they created the ai, the chatbots and copilots.

Rabeeh: Now they have, they're targeting IGI and AI innovators, which will innovate and do discovery. [00:17:00] So we are going towards that path. Now,

Mehmet: I, I'll ask you indeed, about the transfer repair, but you know, you are, uh. Triggering me to ask more and more questions, which is good, I believe, because we are discussing really in a nice way.

Mehmet: So before I go and ask you, you know, what you do with CSP for, you know, the regional clients here, before I jump to this, Because you mentioned something about also the involvement of data science and all this, and I remember, you know, one of the first things they used to teach us in data structure to do lists, right.

Mehmet: And, you know, later on you start to, you know, from Python perspective or like any language, which is like suitable for machine learning or AI. So they try to, to, to, to teach you to do these graphs things you mentioned about, which is the knowledge base. Actually, you teach it how to search basically. And, uh, you know, of course, you know, there are many companies that pioneered and built some products on top of this, but really I'm feeling Rabeeh like.

Mehmet: [00:18:00] Now, I mean, data scientists role also is shifting. Like, it's, it's not like, you know, about just, you know, getting the data and, you know, uh, draw these nice graphs out of it. Because honestly, you know, I, I tried personally with chat GPT and even like with the, with the other tools as well. And, you know, I'm saying like, okay, what's left for the data scientists or what are they going to do moving forward?

Mehmet: If I might ask the question this way.

Rabeeh: Yeah, that's as well a great question. Not just for data scientists, I think for programmers as well, but data scientists as well. So, uh, I think what is going to always stay is the ability the business analyst. Part. So if you tell me, let's say I go back to school and study now, I tell them a business analyst will always be there as long as the humans have to interface with each other.

Rabeeh: So the analyst, he [00:19:00] will use tools and tools will generate. Uh, maybe most of his work and and we created the tool to generate most of his work, uh, to generate the requirement specification document and the functional specification. But you need that person to sit with the client and to listen to their pain and to feed it to the AI engine.

Rabeeh: So I remember my, the professor I worked with, uh, Which I think, you know, as well, maybe he taught you Dr Abbas. I think he thought both of us. So he used to say, like, the future after 30 years and above is that you tell the computer what you want to do. You give a specification and then the computer will decide how to solve it.

Rabeeh: And that was his vision in the language that he'd written. So, how can you specify any problem by the way, that's why the company is called CSP. [00:20:00] Why you can specify it as constraints, variables, and their domain. And this is an AI problem, which is called CSP, Constraint Satisfaction Problem. And this problem can be anything.

Rabeeh: So, if I want to buy a new laptop, I have set of constraints. I need it. 16 GB ram and I need a touch screen and I these are the constraints and the domain are maybe the market that I am buying from which could be Amazon and other players. Uh, so and the solver will take this input and will and will go trying to satisfy your constraints to find you the most.

Rabeeh: Potential possible, uh, solutions and, and, uh, and, uh, and here there are two schools in the market, uh, and, uh, computer science. There's a school saying, okay, you give me a complex problem before I solve it. Let me try to break it into small, solvable problems. And we call this reduction. It's like the [00:21:00] human brain is thinking because some problems are very complex to solve at once.

Rabeeh: So we need to break it so they break it first, reduce it, then they solve it, and I think we're going towards that. This is where AI is now going and it's very interesting because we studied these algorithms and maybe they were, uh, some companies, uh, maybe in the USA and in Europe and very, very, maybe 2 percent of the world were, uh, using AI.

Rabeeh: Doing this kind of thing. Now it's gonna be accessible for everybody. And back to your question, uh, that the scientists will become more of a business analyst, and they have to increase their communication skills, understand the problem, then feed it to the A. I. And the A. I. I. And this happened to us in a government project here.

Rabeeh: There were two models. We had four AI models. Two [00:22:00] were built custom, and two were generated using auto ml. So, and AI wrote them and, uh, and, and we used ai. We just, uh, my team. Understood the problem, created the data set, massaged and cleaned the data, and gave it to the AutoML platform, which it benchmarked more than seven algorithms, and it did the experimentation by itself, and it gave us the result, and we've saved two months of work.

Rabeeh: So, wow, that's where we are. Yeah.

Mehmet: Wow. Rabbi, I mentioned something which I like, although, you know, and I think this is where your strength comes in also as well. So despite maybe if someone tuned in at the beginning of this episode, he would listen to you, you know, you, you, you're the nerd, the guru, right?

Mehmet: But at the same time, what I liked during, you know, the conversation, you mentioned like it's sitting with the customer and understanding. Now let's focus more [00:23:00] on, um, You know the business part a little bit of what you do with csp, you know, and with the regional clients now I know like everyone i'm sure they are running after some ai solutions, right?

Mehmet: Everyone wants to to adopt ai Uh, and I think we are kind of lucky by the way. Rabeeh is also in dubai So both of us we we live in the same city So we know like here at the uae and you know majority of other You You know, GCC countries, you know, they're trying also to get this, uh, edge, I would say of adopting these technologies.

Mehmet: Even now we have a chief AI officer almost, you know, placed everywhere to put things in, I would say context and for people to understand what are like some of the problems people are coming to you for. Solving when it comes to AI. So if I want to know or or like, you know, is there any Common [00:24:00] patterns I would say you have spotted in the last I would say two, three years since the hype started, everyone to adopt AI, LLMs and, and so on.

Mehmet: So what you can tell us about that specifically with your work at CSP?

Rabeeh: Yeah, sure. Thank you. It's a good question. So, uh, since that hype started and, uh, there are many common things, but the most common demand we are getting is, uh, AI copilots, uh, and AI copilots on prem. Okay. And these AI copilots, uh, we, like, we did a POC from a week, like a legal, ai legal, uh, assessor that will assess, uh, for example, uh, certain, uh, legal content for the government.

Rabeeh: Uh, as well, we, uh, we are getting AI copilots for customer support. So the concept of the AI copilot, I think is. Prime now because it can [00:25:00] answer the question for customers on different channels, whether this is as on a website, chat bot or what's up or even within the company to to train the existing employees, even in a call center when somebody calls the client.

Rabeeh: So you have an A. I. Cooperative It's a pilot which is already trained on the customer data and is hosted internally and can answer. However you talk to it or type to it, it can give you answer to help the client. So we are seeing lots of demand on AI copilots. That's one high demand. The other second demand we're getting is AI document generation, uh, which is.

Rabeeh: Generating any documents. So we created the system where you can host it on prim and it can generate any document. Whether this is a RFP. I tell you one example we did at [00:26:00] CSP. So if you tell us you want to build this application and we ask you a questionnaire and within two minutes after we finish asking you Seven questions.

Rabeeh: We generate an RFP that proposal for you. We generate the proposal, which used to take us two to three days to prepare. So I remember we had one client approaches this year, and we gave them two proposals. They gave us the request on Wednesday. We gave them two proposals. And, uh, like the client came to us like nobody gave him this.

Rabeeh: We're elevating the response. We're using AI as a catalyst to allow us to deliver better quality for our clients. And that's, This way I document generator we're using currently now for a government department to help them generate documents internally. Uh, this is the [00:27:00] second thing we had the demand, lots of demand for the third thing is more of the Uh, speech transcription, uh, and recognition and summarizing this.

Rabeeh: So we had now a POC from around the months where we're, uh, we're connecting a speech engine on prem. To a teleconferencing system, converting the whole conference into text, then using an LLM to summarize it at the same time. Uh, we're using a I to know how much, uh, like, uh, this person talked versus that person.

Rabeeh: Person and if the session was completely positive or negative. So we're using here sentiment analysis and we're using an NLP, uh, to process and do this. So, uh, this part we're getting a lot. The other part as well. If I rank them, I can say everything [00:28:00] related to LLM, summarization, document generation. This is number one.

Rabeeh: Second part will be the AI model, uh, custom AI model built by data scientists to do, uh, prediction or classification. One case, uh, we did here for Dubai is, uh, we built four AI models for the government, for the customs, uh, at. Dubai customs to classify half a billion commodity that comes to Dubai every day, which was a humanly impossible to check them.

Rabeeh: So so I is checking the committees and, uh, and it checks it on four dimensions. It checks it. Uh, uh, oh, I cannot talk much , of course, of course. And the a stuff. But, uh, we're using ai, uh, where, uh, where to aid the humans. And I think, uh, you, let's say AI is [00:29:00] going, uh, to a level that will be smarter than any human who ever lived.

Rabeeh: And it'll work 24 7 and. Whatever like as a person can do in one year, maybe with all these quantum CPUs, they will process it in a few days. So, so the danger I think here is how we can constrain AI and use it as a tool and not use it as our replacement. I think this is the social dilemma we're entering into is like we have to keep it as a tool.

Rabeeh: We cannot operate. It will outperform any human, but we have to keep it as it as a tool to help humanity not to replace. I think this is where the ethical part off a I, which governments have to put rules, which companies have to put rules. And I think this is coming now. The market is not The grasping what [00:30:00] I can do, but I think maximum by five years, you're gonna see maybe less.

Rabeeh: I think when a G I is gonna boom and I do everything human ever imagined. I think this is where governments will start to put rules on companies because imagine I can create Uh, 100 A. I. Employees or 1000 A. I. Employees. And this is I think we're once we reach that state governments will put rules.

Mehmet: Yeah.

Mehmet: Um, I guess. Yeah, I don't know, like, if the time would allow us. For me, I'm free. Uh, yeah,

Rabeeh: I'm, I'm free.

Mehmet: Okay, good. No, because I think this is, this is important for, uh, you know, the audience globally and locally, I would say regionally, let's call it this way. So when you mentioned these use cases and what people, so I think the reason why they want the on premise part, it's because of privacy and security concern, right?[00:31:00]

Rabeeh: Yeah, yes, exactly.

Mehmet: So this is nice. So when it comes to two things, like on average, like how long, let's say the deployment of such project would take, like, is it like a one month, two months, usually

Rabeeh: look, uh, usually who's, uh, Asking for these projects, the AI models usually are established companies with a few hundred employees who have processes.

Rabeeh: So the most time it takes us is to extract the business requirements and have all the stakeholders there. So this usually the discovery goes around 10 months. To, to do all the aspects because not all companies are prepared or prepared right? Use cases, uh, after we prepare the use cases, each AI model takes around two months with testing and deployment to production.

Rabeeh: And, uh, the LLMs with the platform we [00:32:00] did. At the C. S. P. It's like a no code platform. Like if you're building a basic a I compile it, then we can get you in less than a month ready to the market. Uh, it all depends on, uh, Uh, like your current existing processes and how you want this A. I. To be connected because some companies tell us I need this A.

Rabeeh: I. To work with my existing system and I need your system to be integrated with single sign on with my software. So just to put a hello world there, stay like two months,

Mehmet: right? So usually, let's talk about Google. The other thing, you know, the outcome. So how long it would take them? Do you think, you know, when they come with such projects to see the ROI?

Mehmet: Because, you know, you gave some examples now, and I'm sure I'm sure there's plenty of projects that you have done. Um, and I can [00:33:00] think across not only government, I can think in any domain. I remember you told me before also about like something related to insurance and Such things When when I am the cfo the one who can assign you the check like yeah When I will see my my roi like how do you after of course delivery, of course, we are We understand there is a implementation phase which might take let's say three to six months whatever and But then I gotta start to see my ROI.

Mehmet: So how long it gonna take with me when till I see this ROI.

Rabeeh: Okay. Uh, there are, uh, some kind of ai, uh, like in the speech and LLM, which can give you a direct quick win. So, so if we are talking about, uh, LLM summarization and, uh, uh, LLM document generation, these, you can witness a quick win. Like, uh, around a month and, uh, one to two [00:34:00] months, you'll, you will see directly the impact because they're going to save direct time for you.

Rabeeh: So let's say a, a month, uh, uh, because they're more like a, like a utility or a catalyst, they're helping your team. Uh, for example, the AI models part. Uh, for example, uh, the one we did for the government, it can take the government, um. Each each client has his own different matrix, but if we're talking on average, I can say for LLM and cognitive part and facial recognition and this, it will be like starting from a month and above.

Rabeeh: But for AI models, I can put around, uh, Let's say maximum three months, two to three months because it is, it's integrated in the process and most of the AI models. There's a [00:35:00] learning or training loop. So you're, you're feeding it and training it like a, like a little baby. So, so it will become evolving in time.

Rabeeh: And, uh, for example, in the customs, if we if the catches one case, that's maybe more than half a million dollars. So it's it's, uh, but it will take them to integrate it into their processes and start to use it. A few months because of the size of the organization.

Mehmet: Indeed. Indeed. Now I want to go back to, you know, it's it's something which I'm excited about.

Mehmet: I remember I'm not a computer scientist. Yes, I studied engineering. That's true. But, uh, The AI always attracted me and fascinated me. I want to go back to something and I want to take your, uh, I would say technical expertise slash business expertise, Ravieh. Um, you, you mentioned about [00:36:00] how, you know, the AI would be able, or actually it is able today to take You know, some complex tasks and break them down into smaller ones.

Mehmet: And now we, you know, and this is what I speculated that that is going to happen very fast with what it's called now agents, right? So, so the AI agents, when I saw the first trials of it, like there was auto GPT and baby AGI and you know, these kinds of, of, uh, community projects, I said, man, this should happen.

Mehmet: I don't know why. But it took, I mean, almost two years after chat GPT went wild, I would say until now we started to see open AI speaking about it. Google with the Gemini 2. 0, they are speaking about it. So what does it mean for us, Rabeeh? You know, like, uh, because you mentioned about how AI will elevate us, but how much these AI [00:37:00] Agents can get really smart, not in doing their own task because they need to go and report to some other agent.

Rabeeh: Yeah, other agent, yeah.

Mehmet: So, so, so how smart they can become to collaborate?

Rabeeh: Yeah. That's a good specific question. So the interesting thing about agents, how they work, uh, they've been in the market for long there, by the way, used and by autos and the elevators. So if you have like, uh, yeah. So, uh, they, they will talk with each other and decide which is like.

Rabeeh: If you're on 60 floor, which elevator to send you so that their concept, how they work, let's say, uh, I think that, there's something coming, uh, soon to the market after the AI copilot, which is the AI worker and the AI employee. So the agents can be, [00:38:00] uh. How it is happening now. The A. I. Co pilot is becoming an assistant. So there's an A. I. Assistant project manager. But let's say there will be an A. I.

Rabeeh: Project manager agent, let's say, and there will be inside the company. Multiple agents, one for, um. Let's take a company like, uh, which company let's take, uh, if I put them in my own company, which is easier for me to imagine. So I have the AI project manager. I have the business analyst. I have the tester, the quality assurance and developers and etc.

Rabeeh: So each if how it works that you give them a task or a goal and they will communicate with each other. To see how they can reach that goal. So, uh, what we have started to see in the market, some new startups trying to [00:39:00] create a, like an agent's platform where they can configure these agents for a company and have them, uh, Work together to solve a common goal.

Rabeeh: So this is where you're going to see agents in business. They were solving complex problems, maybe in scheduling airplanes and train scheduling and with elevators and other. Now they're coming to the business world and everybody could have agents. Within their company and these agents have common goals.

Rabeeh: Maybe I can tell them, for example, I can have an agent for accounting and an agent for sales, for example, and an agent for support and the agent for support will receive a message from the client saying, please send me my account. Account statement, then the agent of support will, uh, [00:40:00] talk with the agent of accounting and tell them, please issue for me the account statement for that client and that, uh, accounting agent will communicate with the ERP, get you that, uh, statement and we'll give it to the support and the support will send it through the email channel.

Rabeeh: So this is how agent are, are not just basic workflow. They are smart. Uh, that that will communicate to achieve a certain goal, and they are they are the next step. So after the copilots, there are the agents agents, and they are all empowered by, uh, by R P. A. So even if you look at. Uh, at, for example, Microsoft AI Copilot.

Rabeeh: It has in it an ability, uh, to run a flow and even I like to talk about the CSP brain shift. So we did the product called it the [00:41:00] brain shift dot AI. And we are doing a agent inside it, and that agent will run a workflow, which a developer will write to achieve a certain task, because every company has a different ERP, every company uses different software.

Rabeeh: So you will put that platform in and you will give it that goal and AI will run. In parallel to your company's team to make sure that your company is delivering the quality and satisfying the compliance. And this is, I think, where all these big companies are trying to achieve now.

Mehmet: Uh, you know, like, it seems like we are speaking in the future, but we are not, I know, because these things are happening now.

Mehmet: So you mentioned something about, uh, Age AGI, I mean, the artistry. Yeah. Yeah. So everyone has a theory about it. I have a lot of guests, you know, [00:42:00] in the AI field and everyone, you know, they say something about it, but, you know, I, I, I tend. You know, to, to believe like the people who are within, you know, deep down into the tech would have a better explanation to us in a, as they say, in layman terms.

Mehmet: So what do they mean by the AGI? Is it like the AGI? Because, and you know, they make it scary. And every time OpenAI would have an announcement, people, they start, Oh, they're going to announce, you know, that the scary thing that's going to destroy the world. So, which I, we know it's not so. Let's eliminate this, this part with, with you, Rabeeh.

Rabeeh: Okay. So, uh, okay. So the, the AGI, you know, it's the humans are scary, not the, not the tool. It's the human ideas, like scare, like the idea to, to hurt. Uh, [00:43:00] our to create some damage is more scary than the tool s O. I think that the problem lies in the humans. And this is where governments have to control, uh, to control what is doing so that the I G.

Rabeeh: I would artificial general intelligence. Uh, it's gonna be achieved. They're trying to create a kind of, uh, I remember Microsoft from a few months signing with some nuclear plant to give it enough power to run all these GPUs. So they're creating like millions of GPUs and teaching them all the knowledge we ever know.

Rabeeh: And this AGI, uh, will, will, uh, We'll use different kind of AI. It's not just the language model behind it. There is the language model with the agent with the innovator with the solver with NLP with speech to text with everything. So it's it's like a comprehensive bouquet of AI, [00:44:00] which has one goal to innovate and to predict outcomes.

Rabeeh: And with this kind of AI, this kind of AGI can maybe decode or break any. Encoding or any kind of encryption because it has the hardware power and it has all the capabilities to brute force and try all the possibilities and it's going to do something which we didn't do before, which is how can you solve something which you don't know about?

Rabeeh: So how can you innovate in an area which you don't? No about so it's as if it's gonna generate knowledge on its own. And this is this is the part how they will use it. So the government and in USA they're investing in it. There, uh, there is a Microsoft in it. There is all these [00:45:00] big companies in it, but the question how it will be used and you can read some kind of documents like, uh, uh, there is a, I think a lawsuit between Elon Musk and, uh, uh, They're trying not to have, uh, open AI go profitable, but open AI, 51 percent of it is profitable, which is Microsoft.

Rabeeh: So it is already, as if for me, I look at it as a more of a Microsoft company. So it's coming profitable and a company, it's no longer a nonprofit organization. And, um, it's how they will use it. If they use it as a tool, To empower people and through subscription that will be good, but the danger if that kind of intelligent will be used in war, it will be used.

Rabeeh: And other things. This is where it will be like, you know, like electricity [00:46:00] or like the Internet. Then there is the A. G. I. There is this intelligence like you're plugging to all these systems, and it's brought forcing all the post. Potential possibility. It's connected to all these, uh, GPUs and maybe some quantum chips, and it uses whatever a I it needs to do.

Rabeeh: It can reduce any problem and solve problems that were never solved before. So there will be a very good part coming from a I and there will be some maybe politician who want to use it for Personal or, uh, some, uh, war related. So I think this is where humanity I, I believe, like, and it's good that you ask this question.

Rabeeh: Like, maybe people like yourself, a mammoth, uh, they need all these guys who have access to this panel, create a kind of community. For a I where they will say [00:47:00] this should not be allowed or this should be constraint and this should not happen. So you need the humanity and the people to come up and put the rules.

Rabeeh: Otherwise, if the business some business guys who don't care about morals and ethics, they will abuse a I everywhere. And of course, the probability of mistakes will happen. And the probability of all these sci fi movies. Could happen. Yeah, I remember there's a philosophical book for somebody who wrote it in the sixties.

Rabeeh: He says that humanity will split into two. One will be very technological extreme. The other will will try to put the rules and regulations and try to contain it. So I think, uh, uh, these maximum 10 years we were going there. People are still not Uh, aware that we're going there, but I [00:48:00] believe no, no, this time we're going there.

Mehmet: No, no, the I always tell people the genie is out of the You know the the bottle It's, it's, it's over and they asked me why I said, okay, look now, let's say even if someone tomorrow decides to close to shut down open AI or shut down Google or shut now humans, they have the knowledge that this is something that can be done.

Mehmet: It's not something fiction that we think, Oh, can a machine do this? And. What also happened? Something magical. I would say the spread off the open source, right? So, for example, which surprises me meta, which is Facebook. They have this llama model, which they made it open source. And I tell people, Hey, how are you going to stop something like this?

Mehmet: Let's say tomorrow we close open AI and we close Google and we, and which will not happen. Of course we know this. So, but there are some people who knows this knowledge now, they know what it can [00:49:00] be done with it. It will not take them. Until they can recreate the same thing because now they know that we have to know how to do it, right?

Mehmet: So this is my my theory

Rabeeh: Yeah, it is. Uh, it is right. It's funny because uh, when we Saw the breakthrough from few years. We started building these two platforms or for ai like this digital brain Which we can put in the company. We started building this and the first step Thing I told the guys, uh, these models are on hugging face and we have more than 10 LLMs deployed.

Rabeeh: Open source, uh, like, uh, LLM, SREL and LAMA 3.2 and LAMA 70 billion and 40 billion and et cetera. So we have multiple LMS available. Uh, open source and, uh, some governments even opened, uh, like in UAE we have the falcon, uh, LLM, which is, uh, [00:50:00] on hugging face. You can download it. Uh, even if you go to this, uh, LLM studio, you can download, uh, up to 10, uh, small LLMs and run them on your CPU and they can answer any question related to like a document that has, uh.

Rabeeh: 20 pages, so you can, uh, You can do that right now in 5 10 minutes. So what we did from two years, the first thing is like, uh, we started creating our own AI so that we thought, like, if there will be constraint on AI as a company, we have all the AI that we need. And we're not the only company who's thinking that way.

Rabeeh: There are maybe millions of people who have done that. Done that way. So I think the genie is, I believe and know the genie is outside the bottle and there is no more. You can't constrain it. The only constraint will come from [00:51:00] governments and they will put the rules on how much you can use a I. But that that's the way it is.

Rabeeh: This will happen little by little. I don't see it happening now. Now it's like the discovery, exploration, automation. So this will happen in the next, uh, three to five years.

Mehmet: Absolutely. And I always tell people, guys, like, this is not a joke. Like you need to really, to wake up now. Uh, because if you think you're going to sleep, still be doing what you're doing today.

Mehmet: My expectation, I'm not talking about AGI. I'm talking about, you know, the way we're going to work. Even if you're not prepared now, after I'm giving it like maximum five years, things are not the same as we are seeing it today because we are on this, uh, extra charged turbo mode now, you know, which is taking us, you know, this very You know, uh, sharp, uh, peak [00:52:00] going up.

Mehmet: No one can stop it. And it's, it's been proven by the day. And this is why, you know, I, I'm not surprised by any breakthrough that's happening because, you know, the, the, just, you know, the fact that now you can do this transformation and because by the way, no one at the show did this. When we spoke about, you know, Chad, GPT and the similar tool.

Mehmet: So of course, we, we talked about the transformer paper, which funny enough, it was, uh, written by people who were working at, uh, Google at that time. So, but we never, no one also highlighted that this is not just this. It's not like just that technology. It's like an LP. It's like, uh, uh, you know,

Rabeeh: yeah,

Mehmet: a lot of things, all the things that we used to see in the AI, uh, 101, I would say, uh, yeah, you know, it's like, it's, it's together and add to [00:53:00] it.

Mehmet: And some people they don't, and thank you for mentioning the RPA by the way, because people, they don't know how much automation there's in the background and they tell them, guess what? You can also automate yourself. Like if you go and you try to

Rabeeh: utilize

Mehmet: an RPA or even a no code, one of the no code automation tools in the market, like Zapier or make.

Mehmet: com, you are missing a lot. And now, you know, last year, really, really, I get tired of telling people. I know some of them, they are saying, no, no, no, this will not benefit us, but really sooner or later, this is going to change. Um, let's see, let's see what the future. So final thing, Rabeeh, like other than the AGI, what, what's, what else is now exciting you, uh, whether it's something related to the company, something related to the technology, like what, what's, uh, Next big thing in your opinion.

Mehmet: Is it the new willow chip from Google? For example, I don't know

Rabeeh: So look like one term [00:54:00] is is changing the whole computation time so this will there are many things good coming and like I have my My sister's child who is studying now in American University, and I told her to do bioinformatics and there will be lots off like what Tesla is doing.

Rabeeh: This kind of hard work that will help. People to see better or to touch, or even they're inventing a synthetic smell. So there's machines now and sensors that can trigger smell, even if you don't have these things. So that will be the future. But everything should be constrained to preserve humanity. We do not want to be lost in AI.

Rabeeh: The next big thing is coming in the health care. And the other, uh, big [00:55:00] things I know it's the age of the LLM. The other, uh, part is the agent and, uh, solvers. So the solvers, which funny were used in the world war two by Alan Turing, who created the Turing machine and all the theory of computation. So, so, uh, the, the algorithm that he did, Was, uh, was that we call it that depth first search, which was basically taking all possibilities and trying to solve and it operates on a concept of a search possibilities.

Rabeeh: And something called a heuristic. And it's funny that we do this every day. So without knowing, so your heuristic is your knowledge. Let's say you have a heuristic to go to the weekend or the holidays and enjoy your time. So that's your heuristic and your search space is maybe these, uh, whatever [00:56:00] available on the internet.

Rabeeh: So heuristic determines everything. It's even in your career. You're a heuristic determines where you will reach. So at each person is like a human solver so that the next part which is emerging now. Now we can move these get these a I solvers that been invented, and this will be the last brain off the robot.

Rabeeh: So now, uh, because for robotics is emerging the a I solver. Which is technically the brain will start to operate because if you look at the brain, okay, you have the vision, which we call this biometric or facial recognition, and you have the inference, the LLM with the knowledge base, what you train, but the solver is what makes you think and search.

Rabeeh: Think and search and, uh, and will guide your search. And this is what will guide, I believe, [00:57:00] the innovator AI innovator, uh, component for, uh, for, uh, open AI. And this is what will guide, uh, will make a thinking machine that will think by itself. So. So I think we're, we're there and, uh, and, uh, the components are just being getting, uh, assembled and the sooner than people will know, they're just gonna see it there and, uh, and see what's happening.

Rabeeh: We at CSP. We're working on that. And we believe we have to two years for only mixing. So imagine options. Open AI with God knows how many more than 10 billion of research with all the brains. So I and they estimated by next year. So I think maximum a year and a half they will hit it. So boom. That's what's the next big thing will be agents and solvers and a I employees.

Rabeeh: So these will be used to [00:58:00] create a I employees and there will be a I employees within the workforce. So the a I copilot came as a copilot to help you. It will mature. But the next step is an AI employee, and we experimented at CSP in an AI assistant project manager, which listens to meetings and then takes the action points and then sets these tasks for the team, and then the project manager can review what that assistant did.

Rabeeh: Uh, and, uh, it will even estimate, uh, how much each task will take based on historic data that we thought it. So it knows if it is building a web application, a form, a form usually takes a day and a half, and so it knows how to estimate the tasks for the employees. And this would been, we did the POC from six months, so, and we are not the only company in the world doing this, [00:59:00] so this is coming.

Rabeeh: So I think, uh. This is this is coming and we started to see it even in Microsoft's new update off the I copilot. So we started to see them going towards that thing. And this is what we're doing. But we're doing it for on Prem and cloud. So for companies who want to protect their data and not have it. On a server somewhere.

Rabeeh: So you can have, uh, our idea is that each company's asset is the knowledge base and its processes. And once you have this system in place, this becomes your system, which allows you to grow. So that's the next big thing coming.

Mehmet: That's great. Rabeeh, I want to make here some takeaways for, for people who still, uh, I would say Underestimate the part of AI and I'm talking here to whether you are listening to us You are in a position of a decision maker or you are even maybe a [01:00:00] technical guy or even anyway So with what Rabeeh mentioned now and all of the discussion we start from the beginning So first if your company is really not adopting AI So you have a problem because someone is outperforming you by minimum.

Mehmet: I would say 10 to 20 times So Right. So this is whether you are from sales perspective, maybe it's a revenue perspective, maybe avoiding risks and avoiding, you know, loss, uh, because, you know, this is what all technology is about. So it may either make something faster, cheaper, or it get you more revenue or it avoid you some risk too much.

Mehmet: Second, this is for like the people in the workforce accepted or not like AI is coming, but not to replace you. to elevate you. So you need to learn how to use these AI tools, because otherwise, you know, to your point about the AI employee. So we need to learn, I think, how to have an AI [01:01:00] colleague, whether in office or in the, you know, remote workspace.

Mehmet: And the most important one, which I take away from you today, is look at, you know, where the technology is heading and don't, you know, Ever underestimate, you know, something and say, no, it's just a trend, a bubble and gonna burst because what we have seen in the past couple of years, especially, we know, and this giant, uh, leap that happened.

Mehmet: So you cannot ignore because now, regardless of the philosophical aspect, I would say of the AGI and, you know, how it should be controlled or not, but as I am seeing, this is, It's a destiny, not a choice for humanity to assume that AGI is here. And in a sense, not of the consciousness and, you know, it's going to go and kill humans.

Mehmet: No, in the sense of it will be a source of knowledge generating force. Let's call it [01:02:00] this way. And it's very, very valuable, Rabeeh. Before we close. Where people can get in touch and learn more about your work and about CSP.

Rabeeh: Okay? So, uh, we are available, uh, you can go to csp uh solutions.com. You can, uh, get in touch with us.

Rabeeh: And we'll be happy to to show you the use cases and how we can help you. And we have many interesting demos we can show and do P. O. C. So C. S. P. Solutions dot com and we're happy to help. Our slogan, by the way, is optimization within your reach because we believe And optimization since 2000 and nine since the inception.

Rabeeh: So I think now it's the age of CSP. Now. Now we believe it. It came to pass it came

Mehmet: in. [01:03:00] Indeed, it's there. So, so, uh, Rabeeh, you know, so for the audience first, the website, you know, will be in the show notes, so you don't have to go search it. I will make your life easy and you can find it in the show notes.

Mehmet: Rabeeh really, you know, I'm biased by, you know, knowing each other's and friends for a long time, but I think this is one of the, you know, rare episodes that we touch on AI. That I think it has a lot of insights and a lot of takeaways I would say and we are not just As we say, you know in our mother language like it's not we are giving theories to people who are giving them actionable Act, you know items So so we were not also talking about something which is science fiction It's something that is here and something they're going to stay so to be really again Thank you very very much for and I know how busy especially it's here and now So, uh, thank you for the time.[01:04:00]

Mehmet: Oh, my pleasure. You

Rabeeh: pleasure.

Mehmet: This is for the audience. If you just discovered this podcast by luck, thank you for passing by. If you like it, subscribe, give us a thumb up and share it with one of the people who keeps coming again and again. Thank you for doing so. I really appreciate it. Send me your feedbacks and comments.

Mehmet: And, you know, I would appreciate also anything that you want me to add, remove from, from the podcast. Thank you very much for tuning in. We'll meet again very soon. Thank you. Bye bye.

Rabeeh: Thank you.