How AI Is Quietly Transforming Pharma, Healthcare and Beyond

Uh, AI agents integrated into their website so the customers can, you know, actually, instead of just filling out a contact us inquiry form, they are actually able to interact through an AI agent and get answers in real time.

The nurse will come in and they'll take more questions and there'll be more data. And then the physician will come in and then there'll be more data. And then when you check out, there'll be more data as you go through reception. I mean, there's a lot of points where they're collecting a lot of information, right? What do they do with all that data?

Yeah, it was uh, basically an algorithm which would identify, you know, these anomalies.

Back then, AI wasn't a buzzword uh, because we were more on the algorithm side. But now, if we can, you know, make it and connect it to a level that uh, it could be qualified as an AI, then we would definitely call that as an AI.

AI, is it hype or help? Good day and welcome to the show. I'm your host, Pete Cooper, and our topic is AI, hype or help? Today's guest is Farz Elahi from Elevate AI Solutions. He's been working in the healthcare and pharma sector and been doing some things to help researchers be more efficient with their time and making their job much easier. So, let's get into it. Farz, welcome to the show. How are you today?

I'm good, Pete. Thank you so much.

Thanks for having me here. Yeah, I'm excited.

Thanks for joining us. Um, we're looking forward to digging in on some of the work that you've done in AI. So, to kick things off, Farz, tell us a little bit about yourself, uh, what you do.

Um, I am a CEO and a chief AI consultant of uh, Elevate AI Solutions. Uh, I founded this company in 2024 uh, with an aim to help uh, businesses grow through AI. Um, as you know, um, there's a huge AI wave going on. Um, so my whole passion was to, you know, help business elevate through AI, maybe through uh, you know, applications or AI agents or whatever they feel like they're lagging in and they can productivize themselves through AI.

So, that's what I'm passionate about and I'm very passionate about um helping uh these particular sectors which I have a deep expertise in which is pharmaceutical and and healthcare. Because I've seen a lot of uh bottlenecks um in those areas where you know, a lot of uh manual work is still done uh which could be productivized and also it could be automated. So, that's been my whole passion um and I also hold a master's degree in data science uh where I get to work along with data and you know, turning them into meaningful insights to be able to impact the business in a meaningful way.

So, first where are you based?

Uh I am based in India. I was in US uh and I founded my company there.

So, I switch between um US and India. So, this uh Elevate AI Solutions is based in USA and it caters across the whole USA and India market.

Nice. So, when you're in the US, where do you operate from?

Uh I operate from uh Indianapolis in Indiana state.

Indianapolis. So, all right.

Yes.

From India and then in Indianapolis. There you go.

Yes. Just a matter of some work.

where you studied, was it?

Uh yeah, I studied in Indiana University. I did my master's in data science from there. I graduated in 2021 from there.

Yeah. So, tell us about I mean, that's a good question that might come up from our audience. What is doing a data science degree?

I mean, when I did my degree so long ago, computer science was just coming in. I I chose electrical engineering but this is way long time ago and then they were just bringing in computer science. Now we have this thing called data science. What is data science?

Yeah, uh I think that's a very good question and I think a lot of people um would want to know, like, you know, what is exactly is this data science, and you know, the different disciplinaries that it follows.

So, consider this as computer science is a huge bucket of items, and then data science is one of the applications of computer science where it focuses more on the data side of things where you could actually have data and do analysis on it. Computer science could also involve you know, building applications, building models and, you know, also some networking analysis that could be done in a whole bucket of computer science, but this data science is a particular segment of computer science which basically focuses on more on data building models, and how we can extract useful information from that data. So, that's how Yeah.

Because we have seen, you know, huge influx of data flowing into the industries and in the market. So, this is where, you know, specialization is there for data science. For people to do.

Again, I'm going to wind the clock back again. Back in the day, before computers were everywhere, like they are now data didn't didn't really exist in the way we think of it today. I mean, yeah, companies would get data from some things that were working and they might even store it in a on a piece of paper, right?

Yeah.

Very hard to access. But now, so much of what we do is online, and so much of our interactions are online that there's enormous amount of collection of data.

So I mean we have a website on ours, and we have a database that's associated with that. So, we have our own database that's associated with our website, collecting data for our stuff. I mean, it's our own blogs and things like that, but it's still a database and there are millions and millions of databases out there and the bigger the company, the bigger the database, right? And so, you guys are looking at all the data that can be collected and saying, "Okay, I want to look at this piece of it for, say, some particular purpose. " So, um I think we've we've ground that one down. So, talk about farmer and health care. So, how did you get into those that area and why you passionate about that?

Yeah, sure. So, right after my graduation, I got a job in one of the biggest pharmaceutical companies in the US, which is Eli Lilly and Company. Uh and I was working there as a clinical data manager. Uh and I was responsible to manage the end-to-end workflows of the phase three and phase four uh Mounjaro trials. And this is where my uh you know, how I developed my interest in dealing with clinical data. Um I solved a variety of problems including um you know, um data in consistencies and making sure the data is clean for uh statisticians and business stakeholders to be able to analyze and make results out of it properly.

Um and I found uh many um you know, cases where I thought uh uh you know, this uh process could be automated because uh there a lot of human interventions was required. Um there was definitely uh a lot of regulations which uh I loved uh dealing with because there are some set of rules which you need to follow and there are some uh set of rules which you can set by yourself uh to be able for the data to be cleaned and you know, um uh be uh ready for the stakeholders. So,

So, what's a good example that people could relate to that you've worked in um you know, I mean, we go to the doctors here in the US and we have to fill out a lot of form, right?

Mhm.

Now, it's usually a tablet or something they give you and you fill out all the stuff. There's all this data going in there.

Mhm.

And then they'll ask then the the nurse will come in and they'll take more questions and there'll be more data. And then the physician will come in and then there'll be more data. And then when you check out, there'll be more data as you go through reception. I mean, there's a lot of points where they're collecting a lot of information, right?

Yeah.

they do with all that data?

Yeah, exactly. I think that's a very good question and with the vast amount of data that they're receiving and it's hard for them to put in the system.

And when what I mean by system is we give them a data entry system where they can, you know, log everything and make sure whatever the data that they're getting is in a digital form. But what happens mainly is because of their busy schedule, they're not able to adapt to that system, data entry system properly. And this is where, you know, they try to or like a kind of enter the data incorrectly. One simple example I could give is the patient was supposed to take four injections and the clinic

Sorry.

The patient was supposed to take four weekly injections.

Oh, injections?

Injections, yeah.

And what Yeah, and what they did was they were supposed to get it those four injections next week, but they got this week. So, the data entry was the problem here in in a way that, you know, the the injections were supposed to be given next week, but they're they were given this week. So, this is where, you know, we would find those anomalies on why the patient was given extra doses. Um And we would we were able to identify those inconsistencies in data through that data entry management system.

So, this is where

So, somebody has some treatment or they have a condition and they have to get some injections, you can look I guess look at similar cases and say, "Oh, well, for that condition or those set of symptoms, they should be getting in these injections in this week. " And and this it stood out as an anomaly, right?

Yeah, exactly.

how you were able to capture that Yeah, that's a good one because it that could really go by and no one would notice, right? And then the person's being in incorrectly administered and then maybe they're not getting better or it's getting worse or whatever because the injections are done out of sequence.

Yeah, and this is how we had, you know, our systems or like data checks established where we could identify whether, you know, the expected amount of treatment was it really received by the patient or was it under delivered or over delivered? We were able to find that through those data checks.

is Is that something you did with AI or is it something you just did with an algorithm?

Yeah, it was basically an algorithm which would identify, uh you know, these anomalies.

Uh Back then, AI wasn't a buzzword because we were uh more on the algorithm side, but now if we can, you know, make it and connect it to a level that it could be qualified as an AI, then we would definitely call that as an AI.

So, I didn't get that. Did you say is AI or it's not AI?

Uh it's not AI for now.

Yeah.

Okay.

Not

So, that was That was done with an algorithm. And so, just for the audience, I mean, a lot of Oh, yeah, throw AI at it. I mean, this is just a word that's a little bit overused, right?

Yeah, yeah.

But you don't always need to have AI. Back in the day, we didn't do anything with AI.

It was all done with software algorithm, which is set of rules, right? If this occurs, then that. Um AI is very good when um there's massive amounts of data

Exactly.

Uh and the rules aren't strict. I think that's probably where I'm going with this. The there's a need there's there's a need for some kind of pattern recognition some kind of um uh inferences I call it in AI. There's an inference under what's going on and then there's perhaps a confidence factor associated with that, but yeah. Okay, so now you're in you're still in this this field of health care farmer. Is there any other examples you've got where you are using AI?

Yeah, so right now in my consulting business I'm using a lot of AI that is helping pharmaceutical companies and hospitals to adopt into these systems. So some of the examples I could give is you know having AI agents integrated into their website so they can so the customers can you know actually instead of just um filling out a contact us inquiry form they're actually able to interact through an AI agent and get answers in real time. This is one of the examples. Um

Like a chat.

Yeah, like a chat exactly like a chatbot and they're able to you know get responses in real time and fast so that doesn't you know make them wait for long.

Other examples include you know me

when people use a chat on a website um what's different between a like ChatGPT is is a big large language model and I think we're just just to clarify for anyone that's listening. When you run a a large language model chat it's looking at the it's using the whole world wide web the whole internet as a source, right?

Exactly.

But when you have a chat on someone's website it will have unique access to information like bookings and you know things specific to the company that it can access. Um so if you want to know more about the company that you're you're dealing with on a website than you're better off to use the local chat, not the not ChatGPT.

Yeah, so

right in saying that?

Yeah, absolutely because the whole point of chatbot integrated into their system or website is basically to restrict the information only within the organization and whatever the customers are interacting to and as you said rightly ChatGPT is a whole you know access has a whole access to all the data around flowing around the world, but this chatbot particularly to that website is only concerned and only restricted and focused on you know how this business can help the customers and how this customers can get information for that particular website. So it's kind of a constraint and trained within the system to

Yeah, yeah.

Well, I think we're all familiar with the chat because what's quite a What what's great about the chat at least this is what I feel is it meets you where you want to be. So if you have a specific question, then you can be deliberate and it will give you a direct answer and then you can say well no, that's not what I want or you can go deeper into it. So it's very much meets you where you are um and that I think is enormously powerful. Before this before these chatbots, you know, you would put a question into Google and it would give you a whole bunch of websites which might answer the question and might not answer the question, right?

And then you would have to go through all those and find the ones that you want. I mean that was better than what we had before that. That's for sure, you know, back in the day you used to have to go to the library. I'd ask my family, my friends they'd go to the library. And that would be a whole morning gone, right? That would be your at school that would be your whole lunch break Like to find an answer to one simple question, which you could literally get in seconds now.

Yeah, exactly.

Yeah.

And also one of the most interesting and my favorite one project that I did with one of the pharmaceutical companies was helping them digitize the scanned documents.

So, what the researchers were doing was using those facts documents which were in paper and in the actual manual files that they were placed. They were trying to look into those and you know, using them to make some inferences for the research to be able to you know, take a a step forward for the drug development process. Trying to see what's the past record of this drug says and the other drug says to be able to make inferences to make other drugs in the market. So, what we did was basically scan those documents. We had those documents scanned. You know, you how these facts documents are. We have a copy of those in a digitized format.

So, we were able to digitize all of them and put it into a structured format so that now they can use computers to actually skim through those documents and also find inferences and make that task easier. So, that was able to reduce a lot of their time. If I had to like give it a number, it could you know, usually save 30 hours of their work per month just to you know, skim through those documents and be able to you know, get it into a structured digitized format was a huge help for them and like it took a lot of burden out of them just to have that system.

This is in pharmaceutical research. Where did all these documents come from?

Yeah, so these were like open source documents that were already released by FDA. So, FDA releases a set of documents of all the drugs that have been approved in the market just for people to see out there how the findings were and you know how effective that the drug was in what particular number of patients. So, those were the open source data that was available to FDA.

brought those documents?

Um, so FDA releases those documents basically.

them, but who actually brought them?

Um, I think it's those internal regulators from the FDA.

Oh, okay.

And and those companies who uh, you know, submit their drug, uh, those uh, findings are uh, submitted by the company.

that don't know the the pharmaceutical industry very well, the FDA is it's a food and drug administration here in the United States. And and uh, and they approve drugs for release onto the market. And in order to do that, they have to do some testing uh, and and verification and it's quite a lengthy process, right? It can take years to get a new drug released. Is that right?

Yeah, exactly. It takes a lot of years.

And then I imagine there's a lot of paperwork because you know, there's submission documents, there's testing documents, there's reports, there's you know, I mean, do you have any idea how many how how big uh, how many documents there are in a drug?

Yeah, I'm sure there are a lot of because like right from subject

hundreds maybe?

Yeah, subject data to like the treatment protocol from submission documents and also like you know, the research paper that goes out after a drug is approved. So, that's like a lot of uh, we are definitely looking at gigabytes of data for just one particular drug.

And so, if just for one drug there's gigabytes of data which is a lot of reading for anybody. And inside all that documentation there might be some useful pieces of information. And I would somebody has to go and find. Um, you know, if you had to do it by paper, you'd you'd have a stack here and you'd work your way through them.

I mean, just with a basic computer, you could certainly collect them all and run searches. But, with a large language model, you can put all of those documents into a file a folder and say just access this folder and talk to it like a chat, like, you know, when was the last um report generated? And in the last report, was there failures of or side effects from the drugs? All those sorts of things. And you can literally just probe the data and it'll it'll go and search it for you.

Exactly. Yeah. That was the whole point and the searchers were really happy to you know, save some of more their time that was taken away by this

Yeah.

manual task and they are very happy with it.

So, did you set up a rag for that a task or what what technique did you use?

Yeah, so basically I used uh OCR technology that was provided by the Meta AI application. And

Optical character recognition. Yeah.

Exactly. Yeah, so that would what this would do is, you know, basically scan those documents and put them into uh different break them into different objects and then those objects were then processed through either um JSON format and then this is how we used to use Python programming language to basically take those pieces of JSON format document and put them into a structurized format.

Right, so um the process was getting a paper document, putting it scanning it.

Mhm.

Then using OCR and if you don't know the audience doesn't know But, I think ChatGPT can do it. Just send it a PDF and it'll it'll give you the text back.

Exactly.

Um So, that's called OCR, optical character recognition. Once you've got that text, then you turn it into this JSON format, which is kind of like a marked-up language, right?

Mhm.

structured better. And then what do you do with it after that? You put it all into a big folder.

Yeah, and then we use a code. Um we build our code in Python programming language to be able to get the JSON file to read the JSON file.

And whatever data format we want, we use the code and to basically

So, I mean, are you using um are you using an a large language model to to access that data or what do you use to actually chat to that data?

Yeah, so the large language model is basically Meta AI that was used for the object character recognition. And also then once the structured data format is done, we would use rag to basically give the users the chatbot option so that they can actually, you know, talk to those documents.

Yeah, so rag stands for retrieval augmented generation. RIG.

And that is where, and just for the audience's sake, that is where you use a large language model like ChatGPT and you use the power of the model as a chat, but it's only looking at that data that you put into that folder or into that database. Um that way it's only can answer data that's in there, but you're using the power of the model as a like a front end. I think of it like a front end for for you. Have I described that well?

Exactly, yeah. You put it in the right way.

So, was that a hard thing to do to roll out a rag like that, a project like that?

Uh initially it It really hard because we didn't have a lot of experience doing it, but as we kind of learned through our way and um uh just to make um you know, everything accessible for our users. Initially, it was hard, but then as more and more documentation from these LLMs came in and you know, we had to do our own research and experimentation. Definitely, it was a lot of experimentation, failures, and you know, looking at the model uh not trying um to do it everything.

Do you remember any of the failure points where you had some difficulties that you you

Yeah, yeah, yeah. Absolutely.

When we were able to build the rag system and give the chatbot to our users for use to interact with those documents. Um sometimes the um chatbot wouldn't respond with the exact information. Um it would, you know, generate some vague results and this is where we found that yeah,

Was it hallucinating?

Exactly. The model was hallucinating and then we had to go back in and see whether all the data is getting passed is getting feed to the model and we identified only 60% of all the data which was in the structured data format was only accessible to the model. So, we had to make sure that you know, every single data point is kind of connected to the chatbot and

That's interesting.

I don't know. I mean so This whole thing about hallucinations, it kind of goes with the term G, right? That the word the letter G stands for generation. The way I understand it is that there's this element of randomization that is injected into these models to force them not to just do the same thing every time, right? Because you don't want the same answer every single time. Everything just comes to the The the whole model won't work if everything comes to one point, right? Everything's pointing to one point. I think that's happening anyway.

Yeah.

And there were also many instances where, you know, we had to tell the model that if you don't know the answer, just tell you don't have access to that information. Because Yeah, because we felt that

making stuff up.

Exactly, because sometimes as people also, you know, when we ask the question, we are feel like we are compelled to answer it even if we don't know it. So, that's how the AI model was to do.

That's right. He wants to play.

Yeah, it has a pressure to answer everything, which isn't always the case because Yeah, we we don't expect AI to be 100% correct, but we at least expect transparency.

Yeah.

with those AI models

That's where it's so important when you're running a rag and you're running it on this data in this profession, you do not want This is one of the cases where you do not want it to do anything random or do anything that's making stuff up. But taking an opposite example, let's say I wanted it to generate a video, wanted to generate an image for, you know, a bit of artwork or something like that that I could maybe even post. That I would want some randomization. I want it to be a little bit different, right?

Exactly.

Yeah, so generation isn't a bad thing. It's just that you need to make sure that you're not doing generation when it comes to something like this.

Yeah, we expect to do the right generation.

Yeah, I mean Wow, I guess you I mean, what do you do when I mean, you had to just solve that problem.

Yeah, so um it was I think going back and forth and seeing um that why the data has been um you know, um like displayed even though it's not available by the model or just trying to identify like why the model is performing so uh differently when the data is not available to it.

So we had to like set up some rules and tell our model that this is you know are the cases that if you go across or you know come across through the data just make sure that you know whatever the right information is you're able to provide that

instead of I mean when you talk about rules we're back to where good old friend the you know the computer science generating rules was what code was built on so you you had to write some rules was that in the prompting side or was that running um are you running a a bit of code in the background how does how did you implement these rules?

Yeah so those those set of rules were as were a part of prompt that we were giving to the model and we're you know identifying the edge cases of when the model could not perform better. So this is where you know we would find those are you know conclusions and we were able to feed back to the system as a prompt and tell them this is how you should perform.

Right. Yeah. So you know standing up a rag relatively easy but getting it to work properly a little bit harder.

Yeah it's a little bit harder you need a lot of experience mentation and going back and forth

How long did it take you to do that project that pharmaceutical documents project?

Yeah so right from scanning to building a chatbot it took us around two months for the whole thing to be completed and then once we were able to have a sample of documents passed through the system we were able to check and make sure that Uh, working for the next fifth, uh, 1, 500 2, 000 documents that's going to keep coming through the system.

So, was that 2 months from start to finish or 2 months and then you then and then you started putting the documents in?

Uh, no. 2 months from start to finish and then, uh, yeah, because, uh, this product was, uh, ready for people to use and then they are still using it to like scan those documents.

So, do you need to stay involved and keep maintaining, um, that system?

Yeah, yeah, exactly because there are some, um, like errors that we do, uh, see. Um, maybe it could be when we upload a file it doesn't go through. So, we had to go back and see why, um, the model is not able to pass the documents. What is the issue? Sometimes it's not in the right, uh, you know, uh, format that we want the file to be.

Yeah.

And then also, like there are some images that doesn't go through because if there is an image it could be a image of image. So, it's not an image but it's a screenshot of an image.

So, it kind of the model has to go like two levels deep, uh, to just pass through whatever is in the picture. So,

do you get to know about the errors? I mean, do do you have a Does the code tell you or do people tell you? How do you know when there's a

Yeah, usually have a once, uh, one meeting in 15 days, uh, bi-weekly meetings with them just to see how their experience has been if they want, uh, to do some enhancement enhancements in the UI or like just get their feedback on how they feel the model should perform and how it's performing. So, it's yeah, it's customer feedback is very very important especially in initial phases of when you, uh, build this AI systems for them.

So, would is this called Did you call this document GPT?

Uh, document GPT, yeah. That was the

that your name?

Uh, yeah, I

You made that up, yeah. That's a good one.

So, if anybody wants to do something like this where they've got a lot of data, they want to put it into a rag, they should contact you and you could do that for them, right?

Yeah, yeah, absolutely. I have my portfolio link that I can also provide to you. Maybe you can put in the description of your video and they can contact me and I can get them set up.

That's great. I think there's definitely going to be a lot of work for you out there. As people try to manage these massive amounts of data.

So, I was talking to a construction person about a year ago. And did you know you might not know this, but there for a typical New York City um um office tower, to build that, the number of documents, I think it would the number was 3. 5 million documents um that needed to be uh created for the building of this um this this skyscraper I suppose or building. I won't even call it a skyscraper. It's a massive amount of data. So, I mean no one person can know all that. So, what they did was they did what you suggested, which is you know, putting basically it's a rag.

Um but the clever part that they created, which is is that it will also scan uh pictures, documents, drawings, everything that has been created. And if you wanted to, for example, check the size of the window opening on level two, northeast wall, you know, you would literally you just send it a question, it comes back with here's here's the size, but also directly referencing work come from the document. So, you'd get the number, right? Let's say 26 in wide or whatever, and then you would click on that and it would pull the document, so you could double-check that it was correctly sourced.

Yeah. Huge benefit for people on the job.

I mean, they use the example of a a junior, you know, project guy who's dealing with a vendor uh and they want to know whether something's included or not. Like, is the cleanup of the gib rock uh what plasterboard, is that included? And the the with the vendor says, "No, it's not. " So, they pull up the the contract. Yes, it is. You got to clean it up, you know. A very practical real-time example um in the construction space.

Yeah, it's very interesting to see, you know, these different applications of AI AI.

Yeah. Throughout the industries. So, it's quite fascinating.

Yeah. So, where do you think AI is going?

Uh I think AI is definitely going in the right direction in terms of, you know, it's uh like if I have to um bet on AI, I would think of something like how, you know, uh the AI revolution is same as the computer revolution back into the 2000s. Like, uh it's going to be same as, you know, you cannot live without computers was back then. Now, you cannot live without AI if you have to be um on the race and uh of, you know, technology, AI is one of the things that's going to take you there and definitely it's going to help um a lot of people and a lot of industries to move forward uh and it's going to be very quick.

Yes.

So, yeah.

But, we have to watch out for the hallucinations, right?

Exactly. Yeah, there are a lot of things that we need to be careful about, which is hallucination as you said. There's also uh regulations. There's also governance uh just to make sure that, you know, everyone who is using AI uh is using for the right reasons and not, you know, doing something uh in the wrong direction, basically. We don't want to people to get harmed by AI and we don't want people to, you know, be scared of AI. Especially, you know, parents have this uh intuition that, uh you know, their kids might get, um uh be, um in the wrong side of using AI.

So, that shouldn't be the case and we have to make sure that, you know, uh everything uh that is being generated or used by used through AI is, um making ourselves uh a better person and make in and is helping uh us make a better society

any examples where AI has been misused?

Um yeah, absolutely. I think there are a lot of examples like, you know, um how the fake images have been created.

You know, there's a lot of um you know, um cases in the social media where, you know, uh many personalities, even like athletes or politicians or, you know, uh you know, the different stars, um uh people fake videos and fake images of them just to, uh you know, um spread misinformation and trying to, like, um damage their personality, which I feel is very, um bad in terms of how people are using AI. That shouldn't be the case and this is where, you know, uh institutions and like governments should set uh clear uh regulations for them and make make sure that we are using it in a legal way. So, that's something, yeah, I'm looking forward to.

Yeah, okay.

But as we're getting to sort of end of our time here, is there any questions or comments you you have for us?

Um I think, um questions uh I have when, um you know, just a a question in terms of, you know, how you feel about getting uh into this AI and, you know, because you have seen um multiple um phases of, uh you know how the innovation has been from computer to internet now to data science and now through AI so how do you feel about it like how has been your experience do you feel fascinated or do you feel scared about it?

So you know I I was there when I was introduced to the internet by a friend of mine I was working at university I had a job at the university but my started my career and he comes in and he says oh this is thing called email and I can send you an email and then maybe a couple of weeks later I signed up to it and then we here would send me an email when he was coming up to visit things like that and that was my first introduction to email but it didn't really impress me until until what used to happen in the day back in the days if you wanted data sheets for electronic parts cuz I was in electronics you had to go and buy the books.

You'd order the books and you'd have the data sheets and they'd go out of date and then you'd need literally you need a whole library of these books and they would always always be going out of date they were expensive they were obviously a lot of wasted paper. We thought there was a better way of doing it with putting everything onto CDs so we were going to invest $20, 000 in these in this machine to read all the CDs with all the data on it. Well the internet came didn't it?

That we never bought that machine we just started looking at the websites where everyone was publishing all this data and that was a bigger hard moment for me and that's exactly data data was then updated by the companies that made the parts that we could access at the moment it was updated and then I would say probably there's been 20 aha moments that have been I mean when MP3s came out and you could listen to music from anywhere around the world and people were collecting all these music you know and I've seen these and I think AI is just like that it's full of a hard moments where I'm going to go, "Oh, wow, we can do that. " And it's certainly in the last 12 months I've had many of those.

The difference with AI is the speed. It's just so fast. I mean, I'm involved in it every day of my life and I still can't get a handle on keeping up. I don't even try. I just learn what I can and go with the flow because things are coming so fast. So, I think it's just like another technology revolution. There will be laggards and there will be innovators. There will be the middle group. The difference is have the speed with which this is rolling out. That's the I guess that's the scary part, too, right? I mean, it's a little out of control. The regulations are going to lag. People's understanding's going to lag. Children are vulnerable to it.

There's lots of little problems that are come up. Mostly because we can't keep up with this uh this incredible speed.

Yeah. I 100% agree with that. The speed is one of the most important

I thought so. I have one last question for you. Um AI, is it hype or help?

Uh I think it's definitely more of a help than a hype. Uh and I explain this in simple terms as um it's definitely helping a lot of people and organizations um productivize and scale their operations and innovate in a very um you know, very fast manner. Uh but the hype it's um generating in the market is uh that it's going to replace humans, which is definitely not the case.

Mhm.

It's going to replace some of the repetitive tasks that humans do, but definitely it's not going to replace humans and um it won't take people's job away, but it's going to make people's job easier and faster and also like interesting. So, this is where I feel the whole world is projecting towards. So, and this is Yeah, and this is very a good time for us to be using AI for the good, for our use, and making sure that, you know, like the innovations and the advancement we make are also in the right direction.

All right. Well, thanks Afraz. Thank you so much for coming on the show. Really appreciate your time, and I'll have to have you back again, too. Thank you.

Yeah, absolutely.

It's been an honor. Thank you. Thank you so much for having me.

Bye-bye.

Thank you. Bye-bye.

Well, I hope you enjoyed that podcast with Afraz. He was talking about healthcare and pharma. On the healthcare side, he's been developing chats for websites that allows patients and newcomers to the website to understand how to interact with the company from the get-go, and that's a very common tool for a lot of companies to have a chat on their website. Secondly, he was talking about how he uses a rag, or retrieval augmented generation tool, to essentially make accessible a very, very large number of documents, and have that easily accessible through a chat, a technique called a rag.

But, he also talked about how that technique can be prone to errors and hallucinations, and what he did to solve that. If you want to know more about Afraz and his business, please follow the link in the description. If you want to know more about Skilful please also do the same. I hope that helps, and thanks for watching.