IGHS64 - Interview with Maged Helmy, CEO at NewCode.ai

Jim Merrifield (00:00.6)
Well, hello and welcome to the InfoGov Hot Seat. I'm your host Jim Merrifield and with me today is Maged Helmy , CEO at NewCode.ai. Welcome Maged how are you?

Maged Helmy (00:10.21)
I'm great thank you Jim for having me here it's a pleasure.

Jim Merrifield (00:13.676)
Yeah, it's great to have you in the Hot Seat! Get to know you and your company a little better. So let's dive into the details here. Can you please provide a brief introduction of yourself, your current role, and one fun fact about yourself?

Maged Helmy (00:24.036)
Yes.

Maged Helmy (00:27.592)
Thank you, Jim. So I am the CEO and founder of NewCode.ai. I'm an associate professor and I completed my PhD in artificial intelligence from the University of Oslo. I've been building enterprise-grade AI for the past eight years for different industries. And then I came across the legal industry, which was document-heavy and process-heavy.

So I started NewCode three years ago, and the target was to improve or to empower the lawyer in their day-to-day operations using generative AI solutions. So one fun fact, we started NewCode at GPT-2. That's not a lot who can say that. So when GPT-3.5 came out, we were like, yes, finally a great innovation. Little did we know today with 4 and 0.1 and Gemini.

how exponential this technology is evolving throughout the months at this rate.

Jim Merrifield (01:31.373)
Wow, that's amazing. GPT-2. Wow, amazing, amazing.

Maged Helmy (01:33.978)
Yeah. Back in The days when it was open source, it's available. Anyone can pull it and start working with it. But OpenAI is now closed AI. So yes, we've been at it. My PhD was very much focused on deploying AI in a secure and encrypted environment.

Jim Merrifield (01:47.105)
Yeah, you've been at this quite a long time.

Maged Helmy (02:01.434)
we were actually focused on developing specific AI algorithms that would help the doctors. And it's such an environment where doctors that would share a wall within the same hospital, within the same department cannot share data. So we thought, OK, well, that's a big gap there. AI is not being utilized. They look through these images. Can we just have an AI deployed on their computer so they can utilize the AI and start working with it?

And that's where the legal industry came in, because that's a very conservative sector. They have the client confidentiality relation. And with the rise of GPT, we had the architecture, we had the system and design, and it was just a question, OK, can we move this to protect law firms, protect their data, protect their processes, while at the same time helping them utilize the potential of AI?

Jim Merrifield (02:57.934)
Interesting. That's amazing. Yeah. So conservative industry for sure, but, and ethically responsible. And we'll get into that. So you have a great background for this, the questions I'm about to ask you on this podcast, but let's, we'll get to that a little later, but what inspired you? I know you touched on this a little bit, but what inspired you to start your company and how has that vision evolved over time?

Maged Helmy (03:04.164)
Ha ha ha.

Maged Helmy (03:24.858)
Yeah, the inspiration here was to try to reduce the manual labor as much as possible. I, when I wrote the started like writing parts of the PhD in 2019, I wrote is like remove the manual labor so that the human can do the more creative work for a professional with many years of degree and experience for them to have to rename files or for them to have to,

read through 200 pages to even get the idea of this. That was the spark of it, is to reduce the manual work as much as possible. Now I learned there's a term for it in the legal industry, and that's reduce unbillable hours. So now I say we're here to reduce unbillable hours, but that's how it kind of evolved. Well, we're starting from the manual work and taking it down there. So the inspiration in this aspect is to have lawyers spend more time with their clients.

To understand what's the issues they're facing and to listen to them. In some cases, I feel lawyers are like a therapy session where you get to vent all the problems that you have. Like, okay, no, don't worry, we're gonna fix it, we're gonna get there and actually spend more time with their clients, understand the clients. And then when it comes to the document, they can just, okay, AI, what's the issue here or how, can you summarize to me the important points or?

Can you organize it for me in a chronological order so I can see what's the issue? And that could be sort of the things that we want, a small piece of the pie that we wanted to focus on. that is one of the major inspirations here is to reduce all the manual work as much as possible and move on to the creative part where we're very good at as humans.

Jim Merrifield (05:18.443)
Yeah, I think that's amazing. I Lawyers don't want to spend, I mean, any of us want to spend time doing administrative work. You know, if we can remove that from our plates, I'm sure that's a beautiful thing. And lawyers, I think, are great at asking questions, right? Because your target audience is lawyers, and they're great at asking questions. I think the secret sauce with the AI, in addition to the back end, of course, is the prompting.

Maged Helmy (05:34.736)
Yes.

Maged Helmy (05:46.58)
It's yeah.

Jim Merrifield (05:48.307)
Lawyers are amazing at prompting and asking questions.

Maged Helmy (05:52.976)
Yes, and structuring it in a very certain way to get some certain information. in that case, the lawyers are still much smarter than the AI when it comes to that because we're in the early phases of AI, but we see a lot of gains. We see massive gains already when it comes to automating a lot of the manual work. But when we move away from, let's say, the current statistical predicting the next word or

predicting the next couple of tokens, there will be more and more exponential gains with the use of the algorithm in the day-to-day work of the lawyers. So yes, it's an industry that's document heavy, standard heavy, and lawyers are good at asking questions. And it was like the perfect combination to start exploring AI in the legal sector.

Jim Merrifield (06:46.174)
Excellent. So what do you see are the biggest challenges and opportunities for AI and even information governance, data management and the tech industry over the next? I don't even want to say five years, maybe three years, because five is just too long.

Maged Helmy (06:58.286)
Yeah.

Too long. there are, I do agree with you, and there are timeless problems in AI that actually has existed since 1950s. Like there are some problems that exist still today and will continue to exist when it comes to the utilization of AI. And the first and most important part, you know, is what happens to my data? Where is my data going? Who is training on my data? When I upload this document, where does it go? What happens at the back there?

And these are very important things is that AI should be utilized in a way where it is 101 % secure. many companies say they are compliant. And in this legal industry, you have to be over-compliant. Being compliant is not enough anymore. You've got to be over-compliant when it comes to designing your own platform or when you're using the AI platform.

The security aspect and the cybersecurity aspects is going to increase more and more. It will go from nice to have to must have. I see more and more law firms are not going to even have the conversation if you don't have your certificates and checks and if you don't have all the compliances in order. So when it comes to the actual... So let's say that we sorted out the security part, which is never 100 % sorted.

because there's always going to be some compromises. If you go to the cloud, yes, you can have access to the latest and greatest, but there's a level of risk there. So we have some law firms that want it on-prem, but if you have it on-prem, you might get the third or fourth best algorithm. So you want to the access to the best. So it's always going to be a trade-off. What is your risk appetite versus how much you can get access to the latest algorithm, for example?

Maged Helmy (08:52.752)
But sometimes the latest algorithm is not what you want. Maybe it's a 3 or 4 % incremental in that case. So unless you're you know, some extreme form or to try the latest one, having already the third or fourth best today on-prem will deliver huge value to the the law firm. So that is one of the challenges is the security and the compliance aspect of it. The second aspect as much as

The AI has been out there and been trained on a lot of data by the bigger companies like OpenAI, Anthropic, and Google. We still see a major gap or a major lack in these algorithms being specialized in the day-to-day tasks of a lawyer. And you can easily see that, you know, it's a West Coast, East Coast, or an English lawyer, or Norwegian, or a Spanish lawyer. We all, all of them are using the same algorithm or the same LLM or large language model out of the box.

So it is very, there's a huge gap to get these LLMs and actually adapting it to the legal workflow for the day-to-day lawyer. And that question is going to be more more prominent is, okay, if every law firm in the US or East coast start using copilot, how can a law firm distinguish itself in terms of competitive advantage? So why should a client come to your law firm, but not that law firm, when both of you are using the same copilot?

And I will see that law firms are going to start distinguishing themselves in terms of competitive advantage by saying, okay, we have a very specialized AI that we've created that is adapted to our workflows. And by the way, we can reduce the unbelievable hours, we can increase the accuracy, and maybe you can come to us with more and more tasks. So the adaptability of the workflows is a second gap that we see today with large language models. And this is why there's a lot of startups.

Being Born. They're trying to bridge the gap from off the shelf large language models to delivering it to the law firm. The third part is the general limitations of these algorithms when it comes to data. Usually, maybe a case will have 1,000 or 2,000 pages all in all. But there is contextualization's limits to the AI. Like, for example, at any given time, they're able to process x amount of pages or y amount of information.

Maged Helmy (11:15.56)
And it gets really bad if you upload the other person's contract and then your contract, and then there's an updated version. And then you ask the AI to do something with it. It's not really able to contextualize. No, I'm on with this, but I'm not with that person. So there's still a lot of development to go. There's still a long way to go. And when I say long, in the age of AI, I mean two years. That's long. That's too long now. But there are still definitely

big gaps that has to be filled when it comes to the legal sector. And this is one of them. How are we able to handle many, many more pages beyond what the AI or the large language model can do without compromising the accuracy or the speed?

Jim Merrifield (12:00.941)
Yeah, it makes perfect sense. I mean, thanks for breaking down those three areas. I think they're so important, right? Each of them have their own risks and value there. So I'm sure you talked a little bit about your team. I'm sure you're not a one person show, right? I'm sure you have developers and salespeople, marketing people, and people constantly innovating. So how do you make sure you talked a little bit about the security certifications, And I totally agree with you. I mean,

Maged Helmy (12:16.316)
Ha

Jim Merrifield (12:29.857)
Probably law firms and companies don't even talk to you if you don't have one of those certifications in place. But how do you make sure that your AI solutions are both innovative and ethically responsible? Because there's a fine balance there. I know you broke it down a little bit there in those three points, but is there anything to add there?

Maged Helmy (12:33.648)
Hmm. Hmm.

Maged Helmy (12:50.572)
A comment on the team. I'm very blessed. I feel I have the best team and I do have the best team and a shout out to each and single person in the team who's really pouring their knowledge and energy into this. We have a very diverse and focused team. We have four PhDs in AI and software. Two professors and associate professor myself and someone with a postdoc.

So we're quite advanced when it came to our education and three of us were doing it in the AI field, starting from 2014 and two of us in 2019. And then the rest of the team are focused on different aspects, whether that's databases, engineering, optimization, DevOps deployment and the likes. And when it comes to the ethics part, we tend to...

We have guardrails in the system and to ensure that the system is not able to generate overly extensive deep fakes or anything that can be used in a wrong way or in a different way than it should have been. But we also, when we do the deployment, let's say on-prem, on the law firm, locked somewhere down in the basement with a key chain and they got three guards or something on it, we also trust that the law firm

and we teach the law firm and educate them about this technology because they're going to have cases where their clients are going to come and say, okay, you know, this has happened and it's nice that the law firm actually has hands-on experience on how dangerous this can be. So we have some guardrails that can be minimized, but in exchange, we show the law firms and we explain to the law firms what does that mean and what can that be used for?

So this technology is very much, it is very much quite powerful. is close to, it is a revolution. And I think five to 10 years down the line, we're gonna look back at, yeah, November, 2022, when ChatGPT came out, we all lived through a revolution, but we didn't realize we lived through the revolution. Because things make sense only in retrospect. So...

Maged Helmy (15:10.324)
yeah, so the ethical part, lawyers know that quite well. Our part is to educate and educate and educate and explain where can this go wrong? How can it go wrong? What can be mitigated? What cannot be mitigated? There are certain things that cannot be mitigated and we're very clear about that early on. We don't want to just build hype or close a deal, but just be clear because we hate to disappoint anyone.

So if they come up and say, you know, but your other competitor said they, you know, they can get 100 % accuracy. And I said, there's no such thing as 100 % in computer science. That has nothing to do with generative AI. It's just computer engineering is never 100%. There is a high degree of probability, but we can't say 100%. And when you start using their terms, we're very careful. So from our side also in ethics is making it crystal clear what you can use this for and what you cannot use this for.

Jim Merrifield (16:09.549)
Excellent. Well, you've definitely intrigued me. I'm going to have to, you know, probably schedule my demo to take a look at this off-line ask you a bunch of different questions.

Maged Helmy (16:15.016)
That would be a pleasure, Jim. I'm looking forward to that. I hope we get to meet you in New York last week of March. I believe it's the 24th of March. And if any of your followers or any of your subscribers have questions, I'm a connection away. So just please feel free to reach out.

Jim Merrifield (16:27.543)
Yeah, I'll see you.

Jim Merrifield (16:38.146)
Yeah, for sure. I'll see you at Legal Tech or Legal Week and whatever they're calling it these days in March. I think there's actually an AI workshop on Monday morning. Yeah, I'll be a part of that. So that'll be fun. So listen, Maged you shared a bunch here. And is there anything else that you'd like to share with the audience before we let you go?

Maged Helmy (16:43.62)
Yeah.

Yeah. Yeah.

Yeah.

Maged Helmy (17:02.288)
I would say that start experimenting with this technology in a safe environment. Get a feel for it. Get a feel on how this can be used. Many of the vendors and many of the online videos might say this is how you should use it, and this is how you shall use it. But we're always so surprised. We have 3,000 users on our systems daily.

And we're always so fascinated by the new use cases that the expert comes in. Because we as engineers think that, OK, here is it. This is what you can do. But then the lawyers turn around and say, oh, by the way, I managed to do this. And by the way, I managed to do that. So this technology is here to stay. Maybe not in the same skin and flavor because it's moving so fast. But the fundamental idea of that, I have a document. This is how I used to do it manually.

can it execute it automatically so I can focus on the other parts. I think fundamentally this is here to stay for at least the next decade.

Jim Merrifield (18:05.815)
Yeah, very good. I mean, it's exciting. It really is. I don't think it'll ever replace, you know, business professionals or lawyers. I mean, you still have to kind of review your work, but it could get you get you close. You know, it can get you very, very close for sure.

Maged Helmy (18:20.986)
Yeah, yeah, we have a point on that. We've been, since the company started three years ago, two of our oldest customers have been on the platform for close, just close to two years now. So we have, with their consent, we have the data to understand how do they interact with the system and what is it that they use. So two years is a lot of data in this, in this case, approximately every day we process a billion, a billion token. And what we've realized is a natural

change in how lawyers work and to put it briefly. So if you think of a line graph or a line chart, a line graph, sorry, and we plot it out. how did you do your work before going on to the NewCode.ai platform? And it was 95 % drafting and then 5 % checking for mistakes and QAing and then sending it over. But then we've seen that 95 % of drafting actually went down to 5%.

And instead of drafting, it's now called fronting. And then the actual 95 % became QA. So lawyers are naturally becoming QA-ing and checking the work of AI, and then based on that, doing multiple iterations to get to the answer, rather than using a lot of brain cycles on drafting. So we have seen a fundamental shift. And I feel it's a silent revolution. I've not seen anyone talk about it yet, maybe because there is no enough data out there.

But the general work of a lawyer will be, if there's educations going out there, is how do we teach the lawyer to QA the output of the AI? And QA-ing the output of the AI is not the same as QA-ing their own work. Because AI will introduce hallucinations. It will introduce bias. It will use certain keywords that are not necessarily focused on your field or that are too far from your field.

Maybe it's used in the East, but not in the West, or maybe used in the East, but not, for example, in the UK. And that's an education part. There's an AI literacy part. And in the EU, actually, they made it mandatory. OK, maybe not. I'm not sure if it's mandatory yet, but it's a requirement if you're a company and your employees are using AI that they actually do an AI literacy course. And this AI literacy course is basically these things. When you QA the AI, it's not like you're QAing your work.

Maged Helmy (20:45.432)
It's a bit different. So you should be aware of these limitations so you have an eye to what you're looking for. But once you master that, it's going to empower you and it's really going to cut down a lot of the unbillable manual boring work.

Jim Merrifield (20:59.874)
Yeah, very good. We'll have to wait and see. It's exciting times for sure. Thanks so much for joining us today on the Hot Seat. It was a pleasure. It was a pleasure talking to you. Yeah, absolutely. And I can't wait to meet you in person. For our audience, if you'd like to be a guest on the InfoGov Hot Seat, just like Maged here, all you have to do is submit your information through our website, infogovhotseat.com. And thank you so much. Enjoy the rest of your day.

Maged Helmy (21:07.704)
Only a pleasure. Only a pleasure, Thank you, Sal.

Maged Helmy (21:26.736)
Thank you.