📒 Download Swarm's free starter guide to fractional work!

Back to Fractional Podcast





Joshua Arvin Lat

Leading AI transformations

Joshua Arvin Lat is the Chief Technology Officer (CTO) of NuWorks Interactive Labs, Inc. He is also an AWS Machine Learning Hero and he has been an international speaker on machine learning, engineering, security, and management.

He’s the author of three books — “Machine Learning with Amazon SageMaker Cookbook," "Machine Learning Engineering on AWS," and "Building and Automating Penetration Testing Labs in the Cloud." He’s also an AI Hive leader on Swarm.


In this episode

In this episode, we talk about how he got to write multiple books, leading as a CTO in large organizations, and how to execute digital and AI transformations with technology.

Outline & Transcript


And at the end of the day, it's about understanding what people need, what your customers need.

So when you're a CTO, it's not just about being technical, it's about having the right solution and team to solve the actual problem.


Directions to show where you can learn from top consultants in AI, software engineering and design.

In this episode, we talked to our Hive leader ARVs.

Joshua Arvin Lat is the Chief Technology Officer of New Works Interactive Labs.

He's an AWS Machine learning hero and has been an international speaker on machine learning, engineering, security and management.


He's the author of three books, Machine Learning with Amazon Sagemaker Cookbook.

Machine Learning Engineering on AWS in building and automating penetration testing labs in the cloud.

In this episode we talk about how he got to write multiple books leading as a CTO in large organizations and how to execute AI transformations with technology.


Thanks for tuning in.

This is your host, Alexis Collado.

Welcome to Fractional.

Happy to be here. I'm Joshua Arvin Lat. People call me Arvs and I am the Chief Technology Officer of New Works Interactive Labs and I'm also an AWS Machine Learning Hero.


They are recognized individuals in the community where we champion several techniques, solutions, the knowledge and experience.

And we shared that to the community as a way of giving back.

So globally there are maybe 30 to 40 machine learning heroes.


Wow. Yes, you're one of the few global machine learning heroes, basically.

Yes, yes. And it's a double S who's choosing who, who's going to be the machine learning heroes. So it's actually an honor to be part of that group.

And I'm able to also speak with other heroes from all over the world and we're able to share our knowledge, we're able to help each other and when they're there are new technologies and like techniques, we're probably the ones able to discuss that much earlier and we're able to share it back to the community.


You're an author of three books. Can you tell us a little bit about the titles of those books?

Yes, yes. So I actually brought my second book. So it's Machine Learning engineering on AWS.

OK. Cool.


So it's a thick book.

Yeah, thick book. It's 560 pages, 500, 2560 pages. My other two books, they are much thicker. So the first they're much thicker. OK, so the first book is, I think 750 pages in terms of length.


So it's around this thick. It's focusing on Sage Maker. So Sage Maker is a machine learning service in AWS where you can build and deploy machine learning models in the cloud.

So before, there wasn't enough documentation to help a lot of individuals use this powerful service.


So I wrote a book to help teams and professionals build and deploy machine learning models and even deep learning models like the ones we see today in just a few hours.

A few hours?

A few hours. OK, so that's the challenge if you know the solution.


If you know the tool, then instead of like trying out a lot of different techniques only to end up not having a solution that's working, you have a lot of templates already working and shared in the book.

And it makes it easier to just deploy your machine learning models in different areas.


Basically like what's the story or journey of just you writing your first book and the subsequent books?

It's actually a good question. A lot of people ask me the same thing. Three years ago, when the pandemic started, we were all forced to work remotely.


OK, so that saved me a lot of time. So it usually takes me one to two hours to commute from my house to the office.

Going back home, it's also two hours plus maybe one to two hours for dinner.

I basically saved five hours per day plus the energy needed to commute.


If you add those hours up, you'll realize that you actually have a lot of time at home.

So let's say that you spend an extra 2 hours at work.

You still have 3 hours plus plus, and you're able to easily build momentum when you want to work on something.


Imagine having an extra year or two or three just writing things you'll be able to make the most out of the time available.

So on my end, when I was given the chance to write the book, so the publisher reached out to me.We brainstorm on what topics we would like to focus on.


Why did you choose the Sagemaker one with Amazon, which is all the possible topics that you can?

Yeah. So in Amazon, as well as in other cloud platforms, AWS, Azure and Google Cloud, there are a lot of services, right?


And as an author, you have an option to whether right on something that focuses on, let's say 10 services or 20 services.

And you can also decide to focus on a single service.

So at that point in time, data scientists have a lot of options and also questions on how to deploy the machine learning models they worked hard for.


Cuz it's usually closeted in their own internal machines, right?

Like you can't easily deploy them to mobile applications or different places.

Yes. So when you are trying to work on machine learning projects, there's usually, I mean simplified 2 parts.


The first one is preparing the machine learning model, using existing data, training it and then producing this sort of artifact or something like an Infinity Stone, an Infinity Stone which can do awesome stuff, something which mimics human intelligence.


So when you have that Infinity Stone, how can other entities or users use that?

So you have to deploy it into its own server.

So that's what they call productionalizing a machine learning model.So at that point in time, it usually takes one week, two weeks or three weeks in order to build a custom solution just to make that happen.


There's this service, there's this managed service called Sagemaker, which basically reduces the amount of time needed to deploy the machine learning models in the cloud.

That's just one of the things you can do in Sagemaker.

Let's say you have a large language model that needs to be fine-tuned.


Instead of setting up the entire system yourself using the raw materials and services, you basically just use Sage Maker and put a few parameters and then you're able to optimally, like solve the requirement.

So if you need to fine tune the model, just put the data, just put a few parameters and then wait for a few hours, maybe a few days, and then the system turns off by itself.


And then the next step would be to you evaluate it and then deploy the model after.

So if you know the tool, you'll be able to do a lot of things in a much shorter period of time.

At the current context, a lot of businesses, founders, enterprises want to utilize AI, right?


And Sagemaker's just one of the tools. But what I've noticed is in your second book, you brought in your topic, right? It's not just about stage maker machine learning engineering on AWS, right? So you chose AWS as your cloud service provider, but it's not just like single service, a single service, right?


So can you talk about the transition and was it like writing the second book versus the first one?

So as an author, of course on my end, I don't want to write the same topic twice or focus on the same topic twice.

So it would be weird for me to write a topic again on Sage Maker and at the same time there are other services which work well with Sage Maker and there will be times where you won't be using Sage Maker depending on the use case.


So again, it's not about using technology for the sake of using it, but rather solving a real business requirement.

So when you're trying to solve a real business requirement, you would have to take a step back and look at what options you have and provide the solution using the different building blocks.


So for example, you have this solution and then you have the other building blocks there, what is the most optimal way to solve the requirement.

So my second book basically solves that by providing the different options. And towards the second-half of the book, I basically prepared a lot of different strategies on how to use those building blocks together.


So what are some examples of those options and like an example strategy?

In machine learning and machine learning engineering, once you've done and built a lot of machine learning models, in practice you'll be working on a lot of machine learning models.


You'll be doing a lot of experiments. You'll be doing a lot of trial and error.

So like your recommendation engine, yes, your AI generation tool and other different other use cases.

The requirements will keep on evolving, it's not just a one time deployment and then the project is done.


The project will continue and it will evolve into something super different, so.

I get to own AI road map AI products like in your AI vertical right?


So the solution there would be #1 to automate the process of building machine learning models.


So that's what they call an ML OPS pipeline.

So the building blocks will be used to build a pipeline where you just upload the data and then automatically the different steps in the machine learning process would run using the different services and building blocks.


And of course you will have the opportunity to review it as part of the manual process.

And then there you go. It's automatically deployed in its own production environment.

So it's about automation, building systems.

What's the main benefit for like businesses and startups in doing this?


It's a super good question because data scientists for example, their time is very limited.

So what you want to do is when they have already solved a certain problem, you want them to solve new problems.

You don't want them to basically do the same thing over and over again like.


Maintaining the thing, yeah.

Maintaining the previous projects.

So after you have solved 1 project, you automate that project using a pipeline and then they will focus on new projects where they have to understand the data because you want them to do things that only humans can do.


And you basically have machines work on things which are repetitive just with the use of a pipeline.

So it's like Amalos, like deploying brains per area, right?

Per per problem.

But then, if you're done with a problem already, and you've constructed the brain to handle that, you want to move on to building your next brain.


Yes, to solve the new problem.Yes, because you don't want them stuck managing those projects manually.

So automation plays a significant part in the lives of data science teams as well as the machine learning engineers.



With the advent of AI, it becomes a very powerful tool and I think this is where your thread book comes in, which is around like penetration testing, right?

Like can you talk a little bit about that?

So my first two books focus on machine learning and machine learning engineering.


But suddenly in my third book I decided to write a book on security.

So a lot of companies want to build amazing things.

A lot of professionals want to use different tools to solve business requirements.

However, when you are a startup and you want to grow fast, you want to do things really fast.


When they say really fast, you'll be changing direction.

Maybe every three to six months you'll be trying out.

That's not your.

That's every other day.


But yeah, yeah, sometimes every week, sometimes every day.

And you'll realize that there will be things that you'll be like compliance and security.


And even if you have already like ensured that you're compliant with different regulations and there's a check in every check box in the security requirements, that doesn't mean that your company and your systems are really secure.


It's like the last thing honestly as a Co-Founder of Swarm, right.

It's one of the last things I think about, right, like cyber security, is our system safe or whatever because we're just not at the scale to just focus on that.


Now if you're focusing on product market fit, needing your core product down, but I think when you get to a stable state and you're able to do ML OPS, deploy different kinds of AI models, right and a lot of vulnerabilities come up.


I agree, I agree.

At the same time, the CTOS and the developers building the system, powering the startup, making that startup solve different types of business requirements.

The developers themselves have no idea what to secure in the 1st place and how to protect the systems from attacks.


There will be a lot of best practices, but the only real way to check if a system is secure is by attacking it right? How would you know if your house is safe? You put locks, right? So what if the attacker has some large machine able to break the door down, right?


It's the bill.

Use AI to destroy your thing?

Yes, essentially. It's possible now, so there's something called 10 Test GPT. It's actually included in my third book, two to three paragraphs mentioning that in the past hackers would utilize different tools and they would have to learn these things maybe in one to two or three years.


And my book is there to help security enthusiasts and cloud engineers improve their security skills.

However, right now if you know how to use AI solutions and you know the process of penetration testing, using the right tools, and even with English sentences or prompts, they would allow you to attack machines automatically.


You just type in this prompt and then you don't have to learn how to use the security tools. So very powerful and scary and. It skips a lot of steps. Friction points to just attack, but then because of that it's easier to do it and destroy a company.


Yes, yes. But of course the reason why I wrote the third book is I wanted the enthusiasts and the lifelong learners, even the ones who want to enter the world of security, cyber security and cloud engineering.


I want them to practice in their own environments because right now a lot of companies or organizations would probably deploy their environments in the cloud. Instead of doing things on premise or in their own data centers, they would probably deploy their systems in the cloud.


The modern software engineering organization, right? And the moment they deploy it in the cloud, developers and engineers probably have no idea how to secure their own systems. And even the security specialist, they don't have the experience needed to audit the resources in the cloud because they never started with cloud engineering.


Some of them probably started started with web development and then proceeded with specializing on security. But there is a lack of resource when it comes to cloud security. So my third book is focused on cloud security and building those lab environments where they can play around safely.


So let's say you want to be a security expert. Do you prove that you're an expert by attacking another organization? That's illegal? You need the contract, right? You need the contract and consultant job, yes.

Yeah, so you start small by practicing in your own lab environment, and then you try to get certifications.


You try to get experience and then you learn from those who are already in the industry for a long time. Practice this thing safely and yeah, read the right books and be with the right people and organizations also.



Thanks for your summary of three of your books. I'm sure there are more to come. Maybe knowing who you are, I kind of want to transition to you being a CEO, right? Like, what does that really mean and how do you lead when it's a super challenging environment, right.


And you're your CEO of NuWorks, a very prominent agency in Philippines. Tell us more about that.

As a CEO of a relatively a large organization, so when I say relatively large, anything more than 20 to 30, the dynamic exchange a bit.


So when you're a CTO of a startup, it's different from being a CTO of a small to medium or big organization. Being a CTO of an 18 man team is different from being a CTO of 150 to 200 man team. The responsibilities are super different.


At NuWorks, I have a lot of responsibilities. There will be profit targets, profit and revenue targets. There will be the most important thing. The most important thing, number one, revenue.

Yes, yes, OK.

And at the same time, even if you are super good on the engineering end, you would have to make sure that that's converted into something which the organization feels for example digital transformation.


So again. That's grown a lot, right? Especially in different corporate contexts. It's like always about digital transformation.And using technology correctly. So the real tech experts are the one who are able to utilize technology to solve real business requirements.


Because at the end of the day, technology solutions and tools, they will change every three to six months. You'll see a lot of AI solutions popping out every week, right? There will be new companies and 3rd party subscriptions you can use to solve different things, but at the end of the day, it's about understanding what people need, what your customers need.


So when you're ACTO, it's not just about being technical, it's about having the right solution and team to solve the actual problem or actual requirements. So in our case for example, the first one is building let's say our own internal systems that speeds up reporting and we need to work on that while solving the other requirements like the the profit and revenue targets.


And at the same time, we have to make sure that we are able to utilize AI properly in our organization.And build our own AI tool. So the first book actually came in handy, because after writing the book on Sage Maker, I realized that one of the more powerful solutions out there when building your own custom solution is actually Sage Maker.


Why did you say that?

AWS would not build another similar solution because it already has Sage Maker. So there's Bedrock, but Bedrock solves a different set of requirements. OK, if your requirements become super custom, then that's the time you use Sage Maker.


If you need the flexibility aspect, then Sage Maker is the solution to go to.

So there's a bit of luck there also because after writing that book, the first book, instead of me trying to spend 3 months to six months building our own internal tool, basically a self hosted LLM solution, it just took probably a few days.


Wow, a few.

Days, yes.

And I saw your PR article, right on, I think Adobo magazine, where you're gonna link to that.It's really amazing moving fast in the AI transformation. It's not just about digital transformation anymore, right?


You mentioned before the episode started that it's all about AI transformations. How do you think organizations and startups can adapt to how fast this whole thing is the first one?

Is understanding the real needs of us people.


It always comes. Back to the problems and the needs a lot of. People like try to understand internal workings of AI when in the 1st place. They should check first how AI can really impact businesses and people, because the needs of the businesses and people more or less stayed the same over the past couple of years.


That's one of the realities that leaders should know or be aware of.

But how do you prioritize across your different needs or different domains, right? Like different problem areas, like how do you think about that? As a CTO?

Yes, it's about having those combo moves combo moves.


Yes, yes. What does that mean? So when I say. Combo moves.

It's like in a game where instead of like trying to solve different problems separately, you can just think of it as one big picture where you utilize a building block to solve multiple requirements at the same time.


So some teams would just throw in money. They would just say, oh, let's just hire 200 people. On my end, I probably just need 10 or 20 and have an output that is comparable to the work of 200 people. And at the same time there should be proper analysis on the economics aspect, especially on the financial end, especially now in the pandemic where clients budgets and even our own budgets are stricter than before.


So once you have that financial knowledge, it's much easier to solve the problem because you'll realize that some solutions are actually not possible. So for example, if you are a smaller company and you want to build your own large language model and you want to compete with the LMS of the giants, it's not gonna happen.


It's not gonna happen.

Right, so on your end and you can be a tech organization with AI as the core offering?

Probably not, yes, but maybe an AI enabled company, yes. And using it the right way, because a lot of organizations just want to be called AI powered company, when in fact it's just there to let the investors know that they're using AI.


Using AI properly needs a lot of research and a lot of analysis because you have to think about a lot of things. The first question is, are you really solving a real problem? Some would just use a tool, a generative AI powered tool, and then they would say over AI powered when in fact AI wasn't really used to solve a real business need, a real customer need so, so.


You and I, we always talk about understanding your problem and solving a real business need, and you said that the first step is understanding the requirements, doing analysis, and then being conscious about your budget, basically of the organization.


What are some quick practical tips you recommend to you know other CTOs and consultants and doing the analysis part?

Listening and understanding the problem. A lot of teams, especially the more technical ones, they are super keen to use specific tools just to solve a problem.


There are some cases where you don't even need to use this high tech tools at all. Maybe a manual process could easily solve super important business need. There are times where semi automation would just do the trick. A lot of organizations would say let's automate the entire process.


However, in some cases 10 to 20% of that process is a bit expensive to automate. So having that right mindset and removing the bias because you just read it in the book or just heard it from someone else, you have to really analyze what really works in your own context or in the problem that you're solving.


Every problem, even though they are similar in nature, it requires a deeper understanding and you have to spend time trying to understand what really works. Depending on the use case, a lot of teams have templates. They would say, oh, your problem is like this, so this is the building, these are the building blocks and that's your solution.


They would try to answer it in five to 10 minutes. At that point, we'd probably be thinking already, how's that even possible? Because you have to really listen to what the clients would say. They would need to tell their story.


They would have to share the different parameters, the context, what are they really trying to achieve? You have to understand a bit of their road map as well and how your solution would fit in their overall road map.


So it's about having that mindset where the requirement comes first and technology is just there as a catalyst to support, to enable, enable to achieve their goals.

Basically, yes. So what I'm hearing is really understand the business process, what are the core drivers and don't have like the having the shiny new thing syndrome like being overly biased towards specific solutions.


Don't be afraid to consider like semi automated solutions that don't really require a ton of AI, right.

Or maybe use AI in a select part of the business process that drives the most results.Yes, that's what I'm getting out of what you say.


Yes, yes. You told us earlier that technology changes every six months, maybe even every week for JavaScript like you said. So how do you say what to focus on and what to specialize on?



So the first one is having the right people around you. Each of those members, even though they have job roles, they're all super different from each other. So understanding each person and their basically their skills, their mindset and their approach to solving things, you need to have a really good understanding of that their behavior also.


So once you have a good understanding of your team, it's much easier for you to delegate specific tasks and have them solve things long term. So what I do usually is I try to solve the requirement or problem myself first.


Once I have tried and experimented different things and have discovered the most optimal way to solve it, that's the time I would delegate it to someone, because sometimes a lot of leaders would just literally relay the requirement to other members and they would say that's delegation.


But in real life, how would you audit the work if you yourself have no idea how to do that in the 1st place? How would you know that it's possible to get it done in 2-2 days for example.

So if a member tells you, oh that's not possible, it's going to take three months.


If you yourself were able to do it in let's say one day, maybe let's not expect them to have the same speed, but at least three days to four days, that should be possible and you can now start creating frameworks.

So as a leader, and when you're working with a larger member of team, a large member or large group, you need to have a framework where the team members would simply follow that framework.


So that's one. And then you need to have a buffer, like a 10 to 20% buffer every week where you're solving a super tough challenge yourself.For example, in an executive committee meeting, there's like a super tough challenge.


There are different options there. The first one is to just discuss it with the team directly and ask for their feedback. That's possible. That's probably one of the common ways to do it.

The standard way, yeah.

You brainstorm different solutions, but if you probably know more about the topic, for example AI transformation, the people you speak with probably have less experience compared to what you have.


So what I do in those scenarios is I research it on my own. I allocate time to really understand how it works, because what you see in the Internet may only be half correct. You have, for example, somebody tells you, oh, it's easy to implement and enforce the usage of AI in organizations.


Right now there's still mixed opinions on how to do things. And even after writing those three books, I know that there's still a lot of things out there which could easily shift the balance and change what's really happening out there.


So on my end, I do my own research and when I'm sure that when I have, let's say, three different like strategies to choose from, that's the time I discussed with the team, so that at least we're a bit more prepared and the conversation would be a bit more guided.


So yeah, so think about the tool or the strategy yourself?

Try to implement it maybe through side projects, do your own research and that's when you focus on operational efficiency, right, like creating frameworks for your team.


And I think that really is the role of the CTO basically and just having the necessary knowledge.Maybe this is kind of like architecture level expertise, right?Just seeing things from a systems point of view, but also being able to implement it yourself when it's needed.


Especially for AI.

If that's the new frontier, you have to be tinkering with it yourself. Basically, that's what you're saying.


So Arvs, this has been a great session, and I just have one last question for you.

So you're a hive leader on Swarm, one of our first.


He's basically the most one of the most prominent ones for future hive leaders and hives on Swarm.

What's your advice for them? The first one is about critical thinking and listening.


A lot of us would probably just want to answer a question straight away. So when, let's say when a client reaches out, probably just listen to them for 5 minutes and then you have some canned answers ready for them.


The good ones would stop talking and they would just literally listen and ask more questions. So once you're able to get a clearer picture of what's happening, the next step is about letting go of the technological aspect first and discussing with the client for example, what solutions they have tried already and what didn't work.


So once you have a clearer picture on that aspect, that's the time you recommend solutions and then you discuss the details with the client.

But again, you need to have that good listening ability in order to get all of that completed properly.


And critical thinking plays an important part of it, because there will be times where you will be tempted to use technology for the sake of using it, when in fact maybe no technology is really needed in the first place. Maybe a spreadsheet would solve the problem, maybe there's really nothing to spend in order to solve the same requirement.


Other members would just use super cool AI powered tools, but in fact maybe spending 5 minutes to write it manually. Maybe human brilliance is more important in some aspects.

So I think that is what I can share with everyone, especially now where there's a lot of buzz when it comes to AI.


But at the end of the day, it's about understanding what are these AI tools for. These AI tools were built for people, by people, so you have to understand the human element and the psychology as well. So listen, have critical thinking.


Understand if the solution even requires AI in the 1st place.

Yes, and may we default to human brilliance first before engaging in anything.

Yeah, they did.


OK, Arvs, it's been great having you on fractional.


Thank you so much.

Thank you.