Amr Awadallah's Journey from Tech Giants to Vectara: Data, Cloud Computing, AI, & Entrepreneurship

Raja:
Hello everyone, and welcome to Future of Data and AI. I’m your host, Raja Iqbal. It is my pleasure to welcome to the podcast. Amr is a serial entrepreneur and a prominent name in cloud computing industry. We’ll talk to you about the journey as a technologist and his work at Yahoo and Cloudera, and his most recent startup, vector. Welcome to the show.

Amr:
It’s good to be here. Okay.

Raja:
So,, let’s start with, going back to, you know, you have been a prominent name in cloud computing industry, but I was in grad school. I mean, I looked up to you as a grad student, and you went to Yahoo when Hadoop was born. Correct. Tell us more about it. You know, Hadoop was born out out of the need for being able to process tons and tons of data.

Raja:
So, how did that need emerging? Tell us more about it.

Amr:
Yeah. So, first I want to tell you briefly how I ended up at Yahoo, how Tim to be at Yahoo at the first place. So, I started a company in and with a co-founder. his name is Di Tran, and, we were one year old, five people, in and when Yahoo acquired us. So, I ended up at Yahoo because of them acquiring my first company, and it was a small acquisition.

Amr:
We were five people, one year old. $9 million was the acquisition value. But that was good for my first one. The product that we had was a comparisonshopping product. And when we were acquiring Yahoo, we became the backend for Yahoo shopping. Yahoo! Shopping was their main property that, calls the web for product information, specs, prices, images, etcetera, etcetera.

Amr:
I spent about, three years and the I was shopping team integrating that technology. But then I switched to our role which was focused on data science and, business intelligence, helping a number of Yahoo products like Yahoo search and others to optimize their performance based on user interaction. we had a very large, business intelligence, data science installation, but it was based on Oracle.

Amr:
So, we leveraging Oracle back then Oracle has a technology called Oracle RAC. And Oracle RAC is the clustered version of Oracle which allowed you to install Oracle on multiple servers. And that was our main data warehouse. It was struggling. It was struggling and giving us lots of pains. And the maintains that we had the first one was scaling just getting pains not just were the queries slow to run and loading the data was becoming a problem.

Amr:
it was taking about a day to load a day workable data. So, you can see how eventually that won’t work, that you’ll be getting more data that you can load. So, that was the number one problem, was the scaling problem. the number two problem was the performance and latency of what is as we were issuing them.

Amr:
Also, that was a big problem, like Some of our, queries, in fact, that I remember we had a query for the monster query. It was a very large SQL query that was merging all of the click information with the impression information and the duping that on a per unique cookie basis. a very simple query, but, execution wise was incredibly taxing.

Amr:
Running that query just for one day worth of data would take five days to finish. The query would take five days to finish. So, obviously that was not going to work. We need Something different. And then the third problem was just the problem of not all questions. We wanted to ask were SQL questions like SQL is great, is a very powerful language.

Amr:
I love SQL, but Some of the more involved data science questions, as you know very well, need to go beyond SQL. You need to do Some programing as part of backend the SQL environment, support CDPs, user defined functions and stored procedures and all that kind of stuff, but still very limiting. It wasn’t very capable in terms of allowing us to go beyond SQL as a way to leverage our data.

Amr:
And then I was lucky while I was running that team at Yahoo, that was a very large team, about like & people responsible for this kind of stuff. There was a separate team within Yahoo! that was, responsible for Yahoo search, and they were also, struggling with scaling the whole search back end, back end. And they have seen these great papers from Google about MapReduce and about the, distributed file storage.

Amr:
Google GFS, was the medium of the technology that the paper was based on and they said, hey, maybe we should, borrow Some ideas from Google and build a stable to do. And they started building Hadoop within their team, for the purpose of Yahoo Surf. Right. That was mainly for indexing the web and creating a rich, PageRank index and, etc., etc. to power your search.

Amr:
Now, I was complaining to all of them, that hey, Mike, what is an article on running very slow. And he said, hey, maybe you should try running, Hadoop for this and see what it does for you. And I’m like, okay, I have this query called the monster query. can you help me run that? And he said, sure.

Amr:
Let me let me do that for you. And he noted that day worth of data. And we ran that monster query, which was taking five days on Oracle and on a small Hadoop cluster. It wasn’t a very, very big Hadoop cluster was like & nodes or Something. The query finished in five minutes. So,, So, five days. It was five minutes.

Amr:
And that was like for me that was a like Bob. Like that was just went on like, okay, this is a new clearly this is a new inflection point for data processing. And I want to finish this for Yahoo, get them up and running with this. But as So,on as I finish doing that, I’m going to be leaving the start the company and all this.

Amr:
And in fact, that’s what happened in & That’s when I left Yahoo and started to go there and.

Raja:
Hadoop was by that time, already released on, as open Source project.

Amr:
Yes. Yes it was. The release is open Source, but it was still very primitive at that time. Right. Was mainly MapReduce. there was a language called pig, which was very cumbersome to work with. there was another language from Facebook called hive, which also, was a bit too mature. There was what it can do. So, there were still lots of things that still need to be done to make the technology mature enough.

Amr:
But it was clear that this technology isn’t the path forward. Like if you really want to scale your data processing and you want to have the flexibility to go beyond SQL, you needed Something like that. That was very, very clear.

Raja:
And then, when you, answered my next question partially. Right. So, you decided to move on. You thought that this is going to be Something big, So, we decided to move on and founded Cloudera shortly after you left Yahoo! yeah, we worked at Yahoo for about eight years. and then you left all.

Amr:
So, I was in and, and then and actually launched first to a VC firm called Accel Partners as an entrepreneur in residence. And that happened because one of my managers at Yahoo had left Yahoo and joined that firm as a, as a VC. So, when he heard, I’m leaving, he said, hey, come work here, research your idea and just give us the give us the luxury of being the first ones to bid on it.

Amr:
When you finished building that idea. And that was a very important inflection point for me, because at Accel, that’s what I got to meet my co-founders for Cloudera. So, I got to meet Jeff Hammer backer, from Facebook and his team at Facebook. They had built hive, So, he was very, very instrumental. And he also, coined the term with digital.

Amr:
He coined the term data scientist. So, the term data scientist that many of us are using today, that comes from Dell was at LinkedIn at that time. And Jeff Howard background was a Facebook at the time. So, I bumped into him and he joined. And then while I and Jeff were forming the idea, at Some point I heard about two other guys.

Amr:
their name is Mike Olson was the founder of sleepy Cat, one of the very first, open Source companies in the world. And he was at Oracle. They were acquired by Oracle, and he had just left Oracle. And then Crystal Buckley, who had just left Google. They also, were thinking of doing Something around Hadoop. And Excel was very instrumental in connecting us with each other.

Amr:
So, we got to know about them. We got to meet with them. And we were literally the only four people in the world thinking about commercializing, this technology. So, we said, hey, it’s better off we join forces and build one company right now as opposed to compete with each other from scratch. we dated each other for a while.

Amr:
Meaning? And that’s one of the advice I always love to give to people is you have to spend a lot of effort in picking your co-founders. Don’t sign up with co-founders right away. You have to date them. You have to go out with them. You have to have arguments with them. You have to learn their mannerisms. You’re going to be spending more time with them than your wife and kids.

Amr:
Like that’s like, that’s So, tough. So, take it as seriously as an engagement for marriage. Like it should be Something that.

Raja:
Yes, it’s, that’s the analogy that I often use even for hiring. actually, I mean, now that people out as if you are, you know, getting married, right? So,, I mean, boldly go to get that.

Amr:
Although I would say that co-founding relationship is So,, critical and it’s very emotional. It’s not just work. It’s not just you’re working there. That’s the perSo,n you are relying on to build the company with you. In the same way that you rely on your spouse to build a family with you. So, yes, I cannot stress how you must learn not only how to work with them, but how to enjoy working with them, because you’re going to be working with them for very, very long hours.

Raja:
That’s the. Yeah. So,, and, when you left, did you have this idea, I believe Red hat was around, at that time. So, Red hat was, you know, Linux, open Source and building a business around it. So, did you have a fairly good idea based on Red hat model? how Cloudera would look like as a company?

Amr:
That’s a very good question. So,, there’s two ways I can answer the question. The arrogant way was, of course, I knew this was going to work and everything was clear. And but the truth of it is Sometimes ignorance is bliss. So, I didn’t know how loud it is. What if I was open Source like So,? I thought it was just a good way of doing stuff.

Amr:
Now, I don’t give a lot of credit to my co-founder, Mike Olson So, Mike Olsonagain, he was the founder of sleepy Cat, which commercialized, Berkeley, Apache BDD, which was one of the main, in memory of that business back then. And he was one of the very first companies that, figured out how to monetize open Source in a healthy, scalable way.

Amr:
he had learned a lot of lesSo,ns from that journey with sleepy cats. Then he helped us at Cloudera to implement that as part of commercializing the Hadoop. And it worked great. So, like if I look at our growth story with Google around how quickly within like 2 or 3 years, we were very, very dominant in the markets, when other companies were doing it more in the closed way, they can take 7 or 8 years before they can establish a standard in the market.

Amr:
that, a lot of credit for that goes towards the open Source angle. But also, Mike also,, bringing in the experience of doing that before.

Raja:
Cloud computing has changed a lot since and. Then, I mean, it’s almost mind numbing. I mean, when you look at how quickly things are changing, in the last 1 or 2 years, there have been just, you know, surprising how quickly things have changed. So, it’s lost you. I mean, you have been, since your time at Yahoo!

Raja:
I mean, around and, and, until now. What would you consider as, Some of the most exciting things or catalysts? it could be tools, technologies, any events, any regulatory changes. people that have really shaped, the baby C cloud computing, computing or internet. And when I say this, it, I mean include, you know, pretty much everything that we see around us, right?

Raja:
AI, machine learning and data analytics. So, do you have any thoughts on that?

Amr:
Yeah. So, we have to give a lot of credit to Amazon like AWS. AbSo,lutely. Was the beacon that shine the light on how important it is to evolve the way that developers work. Right. So, if I go back to my first company, which I built in and, when we were building out that web service, gosh, we had to spend like maybe four months, five months finding servers, buying them, going to the data center, racking them, plugging them into the network, configuring them, installing things.

Amr:
So, we had to spend like 4 or 6 month just to get the server up before we can do any coding, and now actually build the thing that we want to build. And today you can go and plug in your credit card and spend that & bucks and you’re going to have an amazing policy built in and like minutes.

Amr:
Right? So, if I were to put my finger on one thing, I would say it’s the combination of these effects versus Amazon training the like on it’s like because that’s how they were doing things internally and that worked for the rest of the world. But more importantly, if you look at the nexus of why cloud computing really took off, it’s because it significantly improved the efficiency with which we as developers can become productive.

Amr:
Right? They literally got improved by a thousand times. Like it’s that significant of an improvement. Like back then it would take forever to get anything up which limits who can do amazing things to a few companies and maybe very wealthy VCs that can fund these companies to go buy stuff. And but very quickly it now became anybody in the room can build an amazing thing.

Amr:
I think that does it that is the genius of the cloud movement now.

Raja:
And that’s, probably the most defining thing in terms of, the BBC technology. Yeah.

Amr:
and it’s still happening right now. Like, we’re not done with that when I’m done with that journey, like that journey of simplifying, how effective and how fast developers can be is underway right now. It’s still going like the cloud was the first the cloud, microservices and serverless architectures getting up, scaling down and, very user friendly, microservice based APIs, self-healing systems, health routing systems, self-correcting systems, data storage just scales up and scales all by itself, all of these amazing things.

Amr:
That was the beginning. Now we’re seeing the next iteration of that. We’re seeing self coding systems, right? We’re seeing what I can go in and define my app at the very high level. And I see the code for the app coming out to myself. Now just to messenger to make it work. Right. So,, So, it’s going to get even easier.

Amr:
Like I see it within five years from now. Yeah, even my mom was now almost eight years old and never, never wrote a line of other than her life. She might be able to create your own app. That will happen in our lifetime. She will go ahead and say, oh, I want to create this new app. It’s called Snapchat, and you take a picture and the picture disappears after five seconds and the app will come out from the other end.

Amr:
It’s fully scalable, fully secure, reliable, full tolerant, etc. etc..

Raja:
So, the people who have the greatest ideas and execution plan would be the winners, as opposed to the best coders that.

Amr:
I think that will happen everywhere. That’s not just for coding. That will happen and movies that will happen in music that will happen in, and in law. That will happen in manufacturing everywhere. If we don’t learn, how can we leverage the AI to make us ten times, & times more, productive and make us focus on the creative aspects of our jobs versus the mundane, repeatable parts of our jobs?

Amr:
Then we will fall behind. So, that’s, that this is a true statement across all jobs like that is going to be happening over the next five years.

Raja:
So, that is probably a very good segway into your current startup vector. So,, Dennis, I understand what the vector, does. but for a layperSo,n out there who’s listening, what is a vector to do?

Amr:
Yeah. So, I’ll give you, three definitions, embedded in varying levels of difficulty. Meaning the first definition is going to be very, very layman, very high level. you don’t have to be an engineer to understand what we do. the second definition is a bit more, technical. And then the third definition is very, very technical.

Amr:
So, the first definition in terms of what we enable is you to build an AI assistant for any business use case that you have inside the organization. This is really what we’re about. We allow you to upload the data around that application. and then you have a prompt on the other end that allows you to ask questions and get very smart responses around that data.

Amr:
today it’s focused on what we call AI assistants, meaning we are reading the information and giving you back the response. The response could be a document would be an answer would be a, a draft would be a, a webpage, whatever. that’s why we give you back today. And I call that passive, right? It’s passive mode where the AI is simply giving you, information in a new form that you leverage to do work.

Amr:
Yeah. I agents, which is we’re very, very quickly moving towards that. about read right. So, not only am I reading the information, I’m giving you back an answer. I’m performing actions now based on that answer. Right. All I’m or I’m performing actions to get you more information to provide you a better answer. So,, to give you a layman example, that maybe, communicates that definition.

Amr:
think of an app like workday. workday is an app that many, many organizations in the world and leverage to manage HR within the companies. today to take a vacation and workday is very hot. You have to go read the manual to figure out what to put in different forms, because you are clearly going to make a mistake.

Amr:
and it can take you five, six minutes to, to do Something that should be finished in no time. So, as an AI assistant, what will happen is you will have a nice box sitting with you within that application where you can ask, oh, I want to take a vacation, what do I do? And then the AI assistant will tell you exactly the steps you need to go through to complete that task, right?

Amr:
But you have to complete your task yourself. So, that’s what we call AI assistants. And that’s the kind of mode that most organizations are at right now, is enabling these AI assistants that tell us the steps. The future, however, is AI agents. So, the AI agents will do that work for you. So, when I say, how can I take a vacation in that, the chat box, the chat box?

Amr:
So, they respond back instead of telling you how to take it, will respond back and say, I can do that for you. When are you leaving? Say, I’m leaving on this day. When are you coming back and coming back at this day? What’s the reaSo,n why you’re taking a vacation from my paid time off? Thank you very much.

Amr:
Vacation files. Right. So, that’s going to be the future. We’re moving in with all of the apps around. This will be completely working with us. And the same way that we work with each other, meaning just my regular English or, Arabic or Chinese or whatever. Communication. Right. Just say what you want, written or spoken. So, that’s the first definition.

Amr:
So, we are about yeah I assistants and AI agents. Now I’m going to step down. as I said, in terms of definitions. Now if you’re going to be deploying AI assistants and agents within your organization as a business. So, now I’m talking to the business users. One of the key things that business users need to be aware of as to is that these large language models today hallucinate a lot, a lot, right.

Amr:
So, I’m unaided hallucination meaning when you don’t give them the data with the response number as high as &%, meaning &% of the responses you’re getting back have Something in it that’s not factual, that is made up. And that can be very, very dangerous in a business context. Obviously, if you’re a lawyer, with a bank or a doctor, you kind of just can’t have that.

Amr:
There was many stories last year about lawyers actually losing their jobs and getting disbarred because they use chatting with you with publicity about Some activity. So, that’s not the number one problem that we So,lve is we So,lve the accuracy and had a selection problem to give you the confidence that the responses you’d be getting back from the system are correct answers.

Amr:
And we call that trusted Jenny AI. Right. So, Jenny AI. So, the second definition. So, the first definition is AI assistance for business. The second definition is trusted Jenny AI. So, it’s Jenny AI that gives you trust. The responses that you can rely on as a business. And that also, does security, explainability and many of the other aspects that require in any, business that would be separation of technology like this.

Amr:
And then the third definition for the technical folks like you is we are read as a service. We have to triple augmented generation as a service. we give you a very simple API where you can upload your documents on one end of the system, and then on the other end of the system, you can issue any problem to anybody you would like, and the system takes care of everything in the middle for you.

Amr:
Finding the right needles in the haystack, retrieving the needles in the haystack, meaning the right passages and paragraphs and all of the documents you uploaded. Reranking all of these. So, I have another model that ranks all these facts in the right order, and then we pass all of these facts to the generative model, which then produces the response with proper explainability and citations while minimizing hallucinations.

Amr:
And then for added production, we have a final model that’s called the destination detection model, that checks these responses for accuracy and gives you a score for every response. Sending you this response is perfect. You can send this back to your end users. You can file this legal draft. You’re not going to get fired or know this response.

Amr:
You should read it more carefully. Something might be off in there.

Raja:
Yeah, that is great.

Raja:
So, earlier days back in grad school, building a model, a machine learning model, it was, many things that we did, as opposed dissertation. Now it is available maybe as a single library using function call in R or Python or Matlab. Right. So, a lot of this has been due to open Source. And you have been a big proponent of open Source.

Raja:
Your second startup actually was all about open Source. Yeah. So,, do you think that, the company needs that you have closed Source models like OpenAI, or will they be actually leading or it will be more open Sourced? models? So, where do you see this whole thing going?

Amr:
Yeah, I love this question. That’s a great question. So, first I want to highlight open Source is a means to an end. It’s not the end. Right. So, there’s Some people that think open Source I want to do Something because it’s open Source or I want to be an open Source company. No. You try to build a company.

Amr:
That is. But in our case, we’re building AI assistance. Right. And Red hat’s case, we’re building an open system. And then open Source is all cloud data. So, we’re building a big data platform. And open Source is a way that you can do that. And business will run that business in a big way. And the answer to your question, though, is both meaning both proprietary systems and open Source systems over history.

Amr:
I get at least if I look at history and if I take a page out of history, both can exist and both can be successful. Like as long as the proprietary system is, well funded and can keep, innovating and advancing at a high enough rate to compete with the open Source. So, many examples of this in history.

Amr:
Right. So, if you look at, think systems, we have Linux, we have windows. Windows is a very, very dominant opening system. Still, if you look at, the phones we have through our Apple, it was completely proprietary. We had Android open both very, very, very successful. if you look at, my models right now, large language models, we have, GPT, and we have, So,nnet and Blob.

Amr:
But then we also, have llama, and we also, have Arctic from snowflake, and we have Mistral. So, I believe the answer is going to be but you going to have to both open Source options and proprietary options, open Source will tend to when if the proprietary option is not good enough then open Source will win. And that happened in the opening system world.

Amr:
Like an operating system or Linux. Almost all other Unix flavors, like there was who remembers So,laris or who remembers silos OS like those many really, really good or weak systems out there that disappeared after Linux came to be because they couldn’t really exceeded. And So, you either have to have a, technical capability that goes beyond the open Source, or in the case of MicroSo,ft, with windows, you need to have the ecosystem like you highlighted earlier.

Amr:
That was a very valid observation, rather that a strong enough that will maintain you even as open Source tries to compete with you. It makes sense.

Raja:
Yeah, it does, it does.

Amr:
And then for Victoria specifically, I’m a very I’m still a very big believer in open Source as Something you should, use in your business. And that’s Something it’s good you’re giving back to the community. So, we’re balancing that. At Cloudera. We were all open Source. And I’ll tell you, I would never do all open Source again. It’s very hard to run a business successfully at scale when you own open Source, because it becomes very hard to differentiate.

Amr:
And not only do you get big vendors like Amazon jumping in and competing with you, by just grabbing your open Source, but you also, get your customers at Some point. And you. Oh, I just got hired. Five people do this instead of you, right? My, I would rather pay people to do it that way. You as a vendor.

Amr:
So,, my advice is you always have to balance how much open Source you’re doing as a business. And as Victoria, we chose that balance in a certain way. Where we’re releasing Some of our models is open Source, but all of our models. So,, for example, one of our most successful models right now is a model called the Hughes Hallucination detection model, or actually and for short, it’s the number one model right now on modeling face for doing hallucination detection.

Amr:
So, if you go to hugging face and just search for hallucination, you will see that the open Source model come up. And then we leverage that model now to create also, a leaderboard or marketplace which ranks or it became now the industry standard benchmark that every single model comes. They rank themselves and add themselves to the dashboard. So, it’s a leaderboard for what are the top models that hallucinate the least when you’re doing drag, right?

Amr:
So, when you’re doing rank specifically, which model should I try to angle anchor myself to? To minimize hallucinations within my infrastructure? So, I highly recommend for those of you building the systems that I want to minimize hallucinations, to take a look at that leaderboard.

Raja:
I would take a look, you know, because, we are also,, building Something, very similar.

Raja:
It’s, the let’s move a bit toward startups and entrepreneurship. You come spend more time as an entrepreneur, I believe, than you spend in working for big companies. So, tell me why startups fail. What would be the top reaSo,ns for a startup to build people?

Amr:
That’s a number one reaSo,n for why startups, various people. It’s either you have founders that fight with each other or you have, the management team is not the management team, or the founding team is not working, together in a healthy way. And it’s not focus on that. I think to build a great company like the not focus on the right things, they’re maybe very excited about open Source or very excited about.

Amr:
I know we should be excited about the customer and are So,lving the problems. That’s why they should be excited about it. AI is the tool that you’re gonna use to achieve that. open Source is the tool that you’re gonna use that you’ve been. So, it’s the people and their experience and are they focus on the right things.

Amr:
That’s the number one reaSo,n by far. The second reaSo,n I would say is bad luck. Like, even if you’re great, even if you’re amazing, even if you’re aweSome, Sometimes you will have bad luck. You came in at the wrong time. A big competitor comes in and just launches Something that, replaces you overnight like that. That’s bad luck.

Amr:
And that can happen. That can happen. And when these things happen, I don’t count them like I do Some angel investments as well with influencers. If the reaSo,n why the company failed is because of people, then I would think twice before I invest in them again. if the reaSo,n why it’s called is because of bad luck, no, I invest them again &0%.

Amr:
Like, no problem. Because I know that these things happen and it happens to the best of us. Like if you look at Steve Jobs, like Steve Jobs before the iPhone, you tried to make Something called the Newton. And know if you heard about the Newton or you you saw it was a few years, like five years before the iPhone completely failed, completely failed.

Amr:
But Steve Jobs, right. Steve Jobs completely said, why? Because wrong time was the wrong time. The people were not ready for it yet. The LG was not ready for it. It was wrong timing. Great idea. Great team, great markets, bad timing. Right. And the bad timing is not under control Sometimes. So, that’s my long answer. If your question that the number one reaSo,n why companies fail is people.

Raja:
Yeah. So, choose your co-founders carefully.

Amr:
Everybody cares &0.

Raja:
For the first hundred hundred and & employees carefully.

Amr:
Yeah. Now I’m So, I interview the first &0 increase in all of my companies. I did that with the previous one with this one as well. after &0, you can start relaxing that constraint a bit, because now we have a good enough seed that this seed of employees now should hopefully spread the right culture, that attitudes and be at the right bar that you expect.

Amr:
So, that that keeps going on. But, you need that first. Hundreds is So, critical to your success. So, you have to be very, very focused on that team that you’re building and that first hundred. And then you have to be very focused, of course, in your early years, first you have to be focused, which most companies struggle with, that they just they keep throwing spaghetti on the wall everywhere, like, yes, you should be throwing Some spaghetti, but not everywhere.

Amr:
You need to like have Some focus, have Some thesis. Otherwise you’re not gonna have momentum, right? It’s like how you roll the lens and there is rays of light that come in and burn a hole. You want it, you’re only going to burn that wall if you have focus. If you’re not focused, you’re not going to do anything.

Amr:
Yeah, especially against the big companies that have way more money and way more people than you. So, number one is focus is very, very critical. And you have to be paying attention to that. And number two is people and relationships within the people. And then number three is the mission is you need to have a very clear mission that you’re working on this because you love it, because you’re So,lving this problem that everybody cares about as opposed to you just working.

Amr:
So, always start with the why? Why are we doing this? Why is this important? And So, at the thought of our mission is we want to help the world find meaning, right? So, the neural networks, the AI is very good at extracting meaning from things and helping you find that meaning So, you can do Something productive with it.

Amr:
So, that So, that’s that’s the mission that I reSo,nate with. And our theme all reSo,nates with and we believe in it, and that helps us work harder. And it also, helps us guide our search strategy or product strategy towards achieving that mission.

Raja:
Okay. So, what advice would you give to aspiring entrepreneurs? I see I bump into people who are building Genii startups, right? So, because that’s where, there’s a lot of opportunity. So, what would be your advice, to entrepreneurs, maybe it’s a likely first time entrepreneurs who who are just taking the plunge on a journey idea. What would you ask them to do and not do?

Amr:
Yeah. So, the first the first thing I want to say and I’m going to say should plug your ears if you prefer, ignorance as a bliss, you should put your hand to yours. Right now. I’m not listening to this, but, but you need to know the started companies is really, really, really hard. It’s really hard, right?

Amr:
Like many people think starting a company is easy is either going to get funding, it’s going to know start a companies. It’s really hard. we see the press and the media. They only cover the successes. They rarely cover the failures. For every success that you are seeing in the media, for every company is getting funded and growing.

Amr:
There is &0 failures that you haven’t heard of. In fact, I joke Sometimes I’d say is the odds of winning in Vegas if you were to go and gamble, your money in Vegas is better than if you were to start a company, because in Vegas, at least you’ll find out by the end of the day whether you have won or lost with another company, you can be five six years before you find out what happened.

Amr:
So, my number one advice is if you don’t have strong resilience, if you don’t know about yourself, that you are a perSo,n that can take punches and keep standing up again, don’t do a startup. Don’t do a startup. You’re not going to survive. So, that’s the number one advice. But that advice also, correlates with because startups are hard and the odds are against you.

Amr:
You look at the math again will startups failed and succeed? You will need luck. The success even if you want. Amazing. If you are the top student in your school, you have &0 million in funding. If you don’t have luck on your side, you’re not going to succeed. So, how are you going to get luck on your side?

Amr:
I usually say if you’re religious, which I am, that you want to go to your mom and you want to ask your mom to be praying for you. So, my religion is, religion and Islam. We’re strong believers that the prayers from the moms are more likely to be, answered than the prayers from, anybody else.

Amr:
So, that’s number one. If you’re not religious, then ask for karma. Ask for luck, do good things good. Do lots of good deeds So, that the karma comes back your way, because you would need that, right? So, and I give the example of Steve Jobs, the Newton he didn’t have luck. Newton fails completely as a device. iPhone just seven years later changed the entire world, right.

Amr:
Because it was the right time and he had the luck on his side. He was great designer. We thinker with everything, but he had the like when luck was inside the second time. So, that’s a very, very important piece of advice. Second is start with why don’t go build your company when you don’t know why I’m building my company.

Amr:
And if you go build your company because you want to be rich, you’re going to fail. Like an investor. See this? By the way, one of the things investors I feel the guy is pitching me because they want to buy the company, get rich. I’m not going to invest in them. I’m going to go Somewhere else because the chances are people they there’s lots of studies on this, and the investors know that they look for that as a signal.

Amr:
Chances are that these companies fail. They are looking in your eyes. The investors, when they’re talking to you, they look in your eyes to see. Do you have a deep conviction about the problem and why you’re So,lving that problem, and how are we going to So,lve the problem? And once they see that, then they believe in you. And once your team sees that, the team believes in you, and once you have all that positive energy and belief around you, then now the chances of you being one of the ten startups that will succeed versus the nine that’s going to fail is going to be way, way higher.

Amr:
So, always start with Y and don’t get excited about the technology. But technology is simply a tool to So,lve the problem. Get excited about the problem and how I’m going to So,lve that problem. So, that’s very, very important advice. And it’s maybe one of the key reaSo,ns why I see, Some of the startups I work with fail is they lose the sense of why.

Raja:
So, what are you most excited about at Big Data in the next 6 to & months, in terms of technology, in terms of business and the opportunities, that you’re looking forward to, anything that you can share, please.

Amr:
Yeah. Well, I am most excited about these workflows, these antic workflows and these antic workflows. Is this, evolution of, rack systems and that these general AI systems to not just, handle information, as in text documents that you have, but also, have the ability to call APIs. And as they call out these APIs, they can call in into databases, they can call into, lookup systems with information, but they can also, call APIs to take actions and do things on your behalf.

Amr:
So, that evolves them to now be more than just simple knowledge retrieval systems to be true workers working with you to make your business method. And that’s very, very exciting for me.

Raja:
So,, in the next phase, what I will do is I will throw Some rapid fire questions. Thank you. Okay. And, you will be given Some choices, and, you choose one of them, and maybe you can say. Because, I mean, choices may not be as clear. you can 1 to 2 sentences. Why you chose one or the other.

Raja:
you would say. No, I mean, I don’t have an opinion or just, you know, but very quick.

Amr:
Okay. I’ll try to keep it quick as you go. The foundry does podcast, I think I tend to ramble a bit, So, I’ve been quite honest. Okay.

Raja:
So,, out of all the models, the large language models that are out there, allow me 3.1 for Ami or Some other model. Which one will be?

Amr:
Wow. That’s a very tough question. But because the answer is depends, the answer is it depends on what task. If I’m doing a coding task or doing a summarization task or doing a business, reaSo,ning business planning task. So, I’d say dependency. And I say that there’s that. I don’t think.

Raja:
You. Well I’ll point it’s So, there’s no clear answer. Right.

Amr:
So, there’s no one model that rules them all. Yeah. Yeah I’m doing. Yeah.

Raja:
And the questions are I mean, So, the way the choices are, the reaSo,n they are, Austin at fire, it is because the choices are hard. Right. So,, I go to fine tuning.

Amr:
I would both both for sure. Yes. There are Some use cases with which I’ll say for most use cases, rag is the right answer. but there are Some use cases where you’re trying to, leverage not just knowledge as an information, but you’re trying to leverage knowledge as an, either a style of writing, a process, a different kind of engineering process, a different type of legal process, then.

Amr:
Fine tuning would be required to teach the process. But if you’re simply looking at information, rag with In-Context, learning will handle &% of the use cases. Yeah. So, my answer is both. Yeah. Wants to add Something very quickly on it. So, see, I told you I remember the lots and I you just going to keep me.

Raja:
I mean please go for.

Amr:
So, there is one very important, thing to note is it’s not just fine tuning because people will say fine tuning with a large context window. If I have a big enough context window, like my context window is 1 million tokens, meaning I can put all of my data, all of my papers that you’re indexing in the context window.

Amr:
When I ask my question, then I will get better results. Even as context windows get bigger and fine tuning gets bigger, rack systems still have their place because there’s many use cases that require the extreme low latency that the rack system is able to do So, very quickly. We find the needles in the haystack leveraging the vector database.

Amr:
And then these needles are the needles that we pass into the context window of the model. Now that model now was fine tuned also, on my data. Then I might get better answers because of that.

Raja:
Yes. So, what we did for our customer was that we actually for our embedding model, we actually fine tuned on their own data. So, I will, we will is more efficient. Right. So, not really fine tuning for generation side but on the retrieval side. And that actually is substantially improve the results because the embeddings being closer and all of that we you understand how this looks like.

Raja:
So, yeah, I mean rag is probably.

Amr:
Tuning them better is a much.

Raja:
Better. Not actually. because the fascinating thing is that, if you retrieval is good, generation is going to be fine, right. So, most models will work fine. This is Something that, does not come naturally to people because, you know, it’s, lines tend to be the heroes of everything, but it is the retrieval, actually, that matters.

Amr:
AbSo,lutely.

Raja:
Garbage in, garbage out as we keep getting that right. yeah. So, we did that. And, and when we did it, it actually substantially improved, the quality of our retrieval, which in turn improve the quality of generation. Yeah. In terms of, enterprise adoption of change, I what is the biggest barrier? Is it technology? Is it budgets?

Raja:
Is it the culture? Is it skill gap or is it Something else?

Amr:
Fear. The biggest barrier is fear of change. Right. and So, the same thing with Cloudera, with big data, with Hadoop coming in like people were using databases like Oracle, Teradata. we have to convince them to move Something completely new. And the number one barrier is the people having that aptitude and having the resilience and the foresight to we need to make this change.

Amr:
If we don’t make this change in a few years from now, are we going to be toast? And that’s the main barrier. Now, once you cross that barrier, now the barrier immediately after it is, that concern. And Gartner and Forrester with this as well, the concern that businesses there’s three things. Number one, the concern that businesses have around the quality of the results because a lot of their initially implementations had the destinations in them and had big mobility issues.

Amr:
That’s number one. Number two, the security and privacy of the data that you put in these AI systems is a very, very big concern. And then number three is the shortage of skilled people like yourself that know how to work with these systems. So, these are the top three barriers often fear. But I would say fear is the number one barrier.

Amr:
You know change is the number one barrier.

Raja:
The impact of AI on workforce jobs eliminated, displaced or created.

Amr:
So, a very, very good question. The summary of it is, net net net net jobs would be eliminated in massive amounts. Net nets. Let me, expand on that. So, first there will be a lot more jobs. There would be new jobs created because of AI. There will be people able to do jobs they couldn’t do before.

Amr:
Right? There will be Somebody who will be able to be a director as good as Steven Spielberg. And they could have never done that before. So, there will be new jobs created. There is no question about that. that’s, this is one of the last skills that humans have. Right. So, when the manufacturing, revolution came with robotics and automation and machines, we replaced our hands, and now we’re replacing our lines.

Amr:
So, what’s left? So, it doesn’t take a genius to foresee that. No, there will be a lot more jobs lost than created in the long term. And in fact, many of our customers, when we were doing POCs today, like part of their proof of value for why we’re doing this, we’re doing this to improve our efficiency and need less junior people coming in.

Amr:
Or in the case of that manufacturing example I gave you earlier, we don’t need to hire the technicians anymore. The workers can just fix the machines themselves So, jobs will be lost. Like, I don’t want to, sugarcoat that. And governments and all the time is not going to be overnight and would take a long time. But the governments need to start being prepared for doing minimum basic income.

Amr:
We will need to do minimum basic income at Some point, or we need to create jobs for people, you know, you know, in Some countries they have people that just press the elevator button right there as opposed to standard elevator. Ask you which button I thought he going for, and then he press the foot and etc. we have Some of those and say useless job because I can’t press the button for myself.

Amr:
But we do it because we want them to earn Some money. They have no other job to do. So, we need to have universal basic income for sure.

Raja:
Yeah. in terms of, Some of the existing big companies, how they adopted, how they actually leverage this opportunity, the opportunity we have MicroSo,ft, we have Google Meet, Amazon, Apple or any other company that you can name. Which company has actually done a remarkable job and leveraging this to their advantage.

Amr:
I think that’s very clear. There’s one company that has done the most remarkable job and never gets of remember, the name is Nvidia. I think Nvidia and Nvidia abSo,lutely is that company. They have been leveraging this the advantage both in terms of the sales they have been doing for the cards, but also, leveraging the AI internally to new things and new products.

Amr:
And like if you look at the robotics stuff they’re working on, if you look at how they leverage the AI to design you A6 they’ve been working on, it’s truly, truly impressive. So, I, I have to give it to Nvidia as the winner here.

Raja:
The that that that’s very true. And friend of mine was joking actually a few weeks back that right now everyone is losing money except Nvidia. Right. So, like, all the companies are losing money customers on enterprise customers, they are losing money. Now it is only Nvidia that is making money at the moment. And it’s very true.

Raja:
Everyone is investing at the moment.

Amr:
Yeah. No I would say I, I don’t the runner up to it. But yeah, abSo,lutely. With the OpenAI with I mean and is a massive, massive success. That’s not estimate like how amazing is. And it’s being used worldwide. Worldwide. It’s a massive brand. I would put them as runner up immediately.

Raja:
And you don’t consider MicroSo,ft to be because if you asked me MicroSo,ft would be I mean they, they, they way they jumped on this opportunity to leverage this and incorporated this into their existing Google system. I think I thought, I mean, bringing OpenAI to Azure, then Copilot and all of that. I thought that it was pretty.

Raja:
I mean, they leverage it very well, but. Well, yeah, Nvidia definitely is the clear winner. here. Yeah. In terms of, when you go and try to sell Victor, to customers, what is the biggest cost barrier or biggest rather, I should say biggest challenge. You know, we’ve talked about barriers, but biggest challenge in adoption, or using Genii.

Raja:
is it the regulatory requirements? Is is hallucination as a data quality. And maybe, perhaps going back to it, the skill gap, once a company actually, adopts Genii. What is the biggest challenge in being successful? Yeah. using.

Amr:
Vignette. Yeah. So, I have a controversial answer for this one, and it’s a bit longer to extend as well, and I apologize, but the biggest challenge is I’m trying to find the light weight and say it is the not invented years in the not invented years. Meaning? Meaning this. If you look at databases & years ago, if you were to go back in time & years ago and you want to build a database, you had to go and build it yourself.

Amr:
You have to go and get the query planner, the query indexer, the SQL parser, the storage systems. And then you had to glue them together in the right way. And then you built your database and everybody was happy doing that. Love doing that, and So, on. But over time, developers very quickly started to see that. Why are we wasting our time doing that when we can just buy a database from Oracle or whatever, and we’ll finish building the thing that’s going to make a difference for our business.

Amr:
And unfortunately, because of how do you this ran this is, a lot of developers things here. We can just build ourselves. Oh, there’s this launching thing. I can just grab & and grab this Victor DB, grab this model, grab that, generative model, and then build a pipeline myself. And yes, you can build a prototype very easily doing that.

Amr:
Yes. Now get the prototype to work in production with low hallucination, with accurate to high quality results, with security, with no data leakage, with no data privacy issues, with explainability being done correct, with the ability to call APIs to do actions and they start putting their head out. And not to mention, the enterprise readiness in terms of scalability and cost, availability in different regions, disaster recovery, etc., etc..

Amr:
So, we are going through a phase right now and that’s natural. I saw the same things happen with Hadoop. with caldera, where the developers really want to do it themselves. I want to build a pipeline myself, and they haven’t really evolved yet into the next phase, which is, oh, I want to build this business logic application for my business So, I can get my raise and earn my salary.

Amr:
I’m just going to use this So,lution to build this application because that’s what they really need, as opposed to reinventing the wheel from scratch over and over again. And not to mention that they underestimate how hard it is. As you said correctly, during the example you gave with embeddings earlier about how fine tuning the embeddings need you to build a much better system.

Amr:
Most of them don’t know that, they don’t know how to do these things, and they end up building very, very crappy stuff. So, if you ask me, what’s the biggest, friction point right now for not just for big data, but I would say anybody working in the space, in the Rackspace, it’s that it’s a how many developers out there thinking they can do it themselves, build it themselves, fail, which wastes about 4 or 6 months, and then they come to work with you.

Raja:
Yeah. That’s great. last question. startups, what is most important? The idea, the people execution, Something else?

Amr:
People &0% underrepresented. Okay, &0%. People &0% is the most important thing is people. execution comes from people. Ideas come from people. And, if you have a great idea, you have met people. It’s not going to work. you’re not going to have good execution if you don’t have enough people. So, it’s people. Number one is.

Raja:
People and good people, but bad people can actually have done a very good idea and do, well. Yeah. Right.

Amr:
Yeah. Yeah. Yeah. Exactly.

Raja:
Yeah. So,, I’m going to, It was a pleasure talking to you. Thank you So, much for your time. I hope you have

Raja:
a great rest of the day.

Amr:
It’s a it’s a.