Speaker 1: 0:08

Hello, welcome to the Microsoft Community Insights podcast, where we share insights from community experts that you have today in Microsoft. I'm Nicholas and I'll be your host today. In this podcast we will dive into can AI do dishes already, but before we get started, I would remind you to follow us on social media so you never miss an episode and help us reach more amazing people like yourself. Today we have a special guest called Kellen Gorman. Could you please introduce yourself?

Speaker 2: 0:37

Sure, my name is Kellen Gorman. I am known as DBA Kevlar. I've been in the industry for a quarter of a century now. It's been a while. Yeah, I'm mostly known in the Microsoft community as kind of the weird Oracle girl that they adopted. I started out doing Oracle on Azure about eight years ago and ended up doing that for over five years at Microsoft before leading up a number of other teams and working with companies as we migrated a lot of these high IO workloads into Azure. But while I was doing that, they found, if I was doing a lot of high IO and all these huge data platforms that I was bringing in, that I also might do pretty well in AI. So ended up kind of becoming an accidental AI technologist. I think that's the best way of putting it. So that's how I ended up in both realms. Got the luck of working in a couple of AI projects recently.

Speaker 1: 1:37

Okay, so why is it called a technologist? Is it because it makes your database that you used to have a background with, and then your new role to AI?

Speaker 2: 1:47

Yeah, you know, ai is only as good as the data that feeds it.

Speaker 2: 1:51

A lot of times, you know we'll come up with this saying or we'll say you know, garbage in, garbage out, yeah.

Speaker 2: 1:56

So it's extremely crucial, as I see a lot of these AI projects those that I've worked on as an engineer as well as those that I may have architected that we find out that development's a little bit different on AI than it is in traditional kind of multi-tier client server environments and things like that.

Speaker 2: 2:15

That you have to think about these a little bit broader and a little differently than you did in those archaic systems. As well as making sure that we're getting the correct data in there, so that you know we get into this. Are you poisoning your data models? We've seen this with a couple of customers, so this has been kind of part of this. So I have a tendency to kind of broaden my reach and saying I'm not a data scientist, I am more of an AI technologist because I've worked in these AI projects in multiple different roles, no matter if it's more of a machine learning when we start talking about synthetic data generation versus when we talk about AI and actually using closest things maybe facial recognition software to identify different species and things like that. All of these projects are extremely crucial to me and important to me, but they require different skills, for me as well as a technologist.

Speaker 1: 3:10

Okay. So before we dive into that, how did you, as yourself, get into AI? Fall into AI.

Speaker 2: 3:17

It started out pretty early on. I have a unique skill set, even when I was at Microsoft, of being able to write everything in Bash scripts. Not a lot of Bash skill set inside Microsoft, more PowerShell, but I had one large customer that needed to convert all their PowerShell into Bash. So what we ended up doing was building a vector file with all of these Bash kind of translations for PowerShell and we utilize that to do those conversions, because there was only one of me and there were a lot of scripts that had to have this done. It was my first introduction as we were using it was with Azure OpenAI services to be able to build this when we first started out.

Speaker 1: 3:59

Yeah, you had to. To PowerShell.

Speaker 2: 4:01

Convert Bash to PowerShell, the other way we actually went from PowerShell to Bash and a lot of folks are like, oh, it's all about PowerShell. Everything is really a data ecosystem, and that was one of the reasons that I ended up coming into Microsoft. I came in to do what ended up becoming Fabric. Later on, within four months, it became very apparent to me that I needed to move the Oracle databases for Microsoft. That was my calling. I said you really need me to move those because everything needs to move into the cloud and Oracle's like holding everything back, and it became a global role for me, and so as we started to build these out, you recognize that you needed more to do on the Linux servers. With Linux, yes, powershell was capable there, but they needed these bash scripts. So that was the first AI project that I became part of.

Speaker 1: 4:48

Okay, so let's dive into the first AI project that you've done. So what model did you use and what do you make it so special to your first AI project?

Speaker 2: 5:03

Oh, can I even remember? We built it. This is early on. We weren't sure it was even going to be able to do this. I mean, we really did. We built out this vector file and just pumped it in through. We were using ChatGPT 3.0 at that point. This is early, early 2023, when we were doing that 22 or 23?, 23, early 23 that we were doing it. So I'm going to say it's January 2023 that we were building out this vector file and doing it, weren't sure it would be successful. It did it about 90%, 90% of the way. And then I was able to come back in and complete whatever conversions were done on those scripts that allowed me to kind of scale up and be able to complete that project on my own.

Speaker 1: 5:49

Okay, so what's the end solution with that project? Was it like a chatbot or was it a different solution?

Speaker 2: 5:57

It really was generative AI. I mean, that's all it was was being able to take those files and convert them. For us to take that vector information and provide that solution for this customer Okay.

Speaker 1: 6:09

Okay, so it's just pretty much like turn to AI you input data and you get data back on what you want. Yeah.

Speaker 2: 6:17

Yeah, and you know we talk about. You know I said why can't AI do the dishes already? One of the things that was interesting to me is that I ended up being part of another project with the London Zoological Society and I'm not going to say where I did it or anything else, because I'll admit to being multi-cloud and they were using facial recognition software through, you know, spectrogram is what it's called with sound that they're able to identify animals ended up going from eight species to 12. And we were using, you know, postgres SQL behind the scenes We'll just say Postgres behind it, and they were able to generate and track all of these animals. One was to remove some of the human interaction with these animals, which both put people as well as the animals in danger, and once we built this all out as the animals in danger, and once we built this all out, I was thinking that this was really great use of AI towards the betterment of humanity and the planet and everything else.

Speaker 2: 7:12

As I started seeing some of the other AI things that were coming about, you would say do you need another chatbot? I think everybody went through this in 2023, where there was an initiative, there was a goal in every product, in every project, that you had to have AI as part of it, and that question wasn't being asked. Does AI need to be part of this? Are we using AI correctly? Are we using AI for the betterment of man? Are we using it? And it comes back to a social media post that a woman by the name of Joanne posted that she says I want AI to do my laundry and my dishes so that I can do art and writing, not the other way around. So that's where this title came from. It was based off of her social media post and my investigation of, like the projects that I was on that.

Speaker 2: 8:01

I really felt good about what I had done with AI and some of the other AI solutions that came out that you kind of said could we have put the GPUs to better use? Could we have done something else? No matter if it was, and we're starting to really evolve our understanding of AI. You know it's not a creature, it doesn't have consciousness yet, but to say this is to protect humanity, or this is to protect the planet, or this is to protect, you know, governments and data and all of this versus. I just created another chatbot that comes out and does a song done in Drake lyrics or something you know, something positive from it versus something just saying I was able to do this, something positive from it versus something just saying I was able to do this. And that brought a lot of value to me, I think, is feeling as a technologist. I was being productive, I was doing something for the better of everyone, not just to create more revenue.

Speaker 1: 8:55

Yeah, because you need to deliver a solution to have a greater impact. So, whether it's all the different projects, which one have the great impact and influence the world, so whether it's the one that goes through the zoo.

Speaker 2: 9:08

Yeah, the one for the Zoological Society. They have now broadened that project so that they are protecting 600 species throughout the UK. There's other ones that are like FitTrack that one's in the US and they're actually tracking with photo recognition software, just like you used to have trackers in the 1800s that would look at a footprint of an animal and say this is a wolf. Crackers in the 1800s that would look at a footprint of an animal and say this is a wolf and I can tell by what they're doing here they're going this direction, they're eating these kind of animals. They're able to do that with just track sightings and taking water level information by the amount of water, knowing where animals are going to. There's another one that is by the. I'm trying to remember which group it is, I will get it for you. But fish, the sides of fish, the spots like on a trout, are unique, like fingerprints, and they're able to take these pictures, these images, and know where trout are going through the locks, through the rivers, like I live on the Columbia River in Oregon.

Speaker 1: 10:09

They probably use machine learning to prediction where the next trail will go or where they will go.

Speaker 2: 10:12

Yeah, so they probably use like machine learning to prediction where the next trail will go or where they will go yeah and then figure out you know, are there challenges to trout getting to their breeding grounds? You know, are we overfishing Stuff like that, that kind of information. In Bristol they're doing all this work with drones, where drones are able first to detect forest fires, all this work with drones where drones are able first to detect forest fires. Two, they can actually do swarms of these small drones and put out small forest fires, so you don't have to worry about firemen being put at risk.

Speaker 1: 10:41

These are the kind of AI projects that I'm very interested in that you are seeing for the betterment of man, the betterment of the world.

Speaker 2: 11:13

Okay, so, from all those AI projects that you've been doing, what's your favorite one? And your favorite? That it's going to help doctors be able to do more with less, but that a AI model may be able to see the risks for a certain patient and be able to help ensure that patients are less at risk as they go through their medical procedures or hospital stays. That is really cool. There's also another AI model that's able to look at healthcare institutions and see where resources are being allocated not just through doctors and nurses, but with medical equipment and realign it so that each process and resource is in the right area so that patients get the best care possible.

Speaker 1: 11:49

Yeah, because I think I read it somewhere that they're trying to use AI to predict to help defeat cancer or something. Yes, with someone. Yeah, so that's. That would be amazing if they tried to fix that Absolutely.

Speaker 2: 12:03

Absolutely yeah. I think it's as somebody who's working in AI or somebody even in general in technology, because we all have like a piece of AI that we're bringing in as well as machine learning. Now we've got AI agents. So a piece of AI that we're bringing in as well as machine learning. Now we've got AI agents. So it's not just that we're returning information, we're expected to do something, respond to it. And as these AI agents are being brought to us by our leadership, by our decision makers that say we need to have an AI agent as part of our goals for this year, it means that it's even more important for us to decide as we bring back this information, as we do something with it, are we providing actual value to the user? Are we bringing actual value to an organization? And I think that's really fascinating, as we start to kind of rewrite and rethink and maybe pull in some of our resources and do it a little differently than before, where we just threw AI out there everywhere.

Speaker 1: 12:57

Yeah, I think a lot of things, since AI is like a buzzword everyone wants in. So now this year, everyone wants in on the agent, the new agent service, oh yeah, so in your current place, what's your favorite AI project that you're working on? You don't have to disclose any details if you want just one of your favorites.

Speaker 2: 13:16

Won't give too much. It's in private preview right now, but it's synthetic data generation and it's ML, of course, and it's using a combination, because we did a lot of research on this and a lot of work on it. I work at Redgate Software as one of my roles, because I don't believe I should just have one job is one of my roles because I don't believe I should just have one job, and with this we're using rules-based data generation, synthetic data generation, along with machine learning, and we're able to do this on just a laptop. I think that's really impressive, because everybody automatically assumes you need to have GPUs, you need to have some really big system to do a lot of this work, and that we're able to do it just on a laptop means that you can kind of reallocate those GPUs or other machine learning capabilities and projects that you're doing, or resources and cost savings there, apply it to something that you really need that power to go to, and this way we can protect a lot of PII, a lot of data.

Speaker 2: 14:11

It doesn't matter if you're talking about FinTech, healthcare, any of these kinds of customers that are concerned about this. Target is usually the best and most well-known Breach, always a concern. Folks know they had to get a new Target card. They knew that there was a breach in Target that most of them don't know. No one ever got into Target. It was actually their test database that went to an HVAC and that was where the production data is. So this idea of synthetic data generation, using ML to generate that and to get millions and millions of records and have that referential integrity there and not have to worry about your data ever being vulnerable that's for an organization or a customer that's kind of my really keen project right now for an organization or a customer.

Speaker 1: 14:55

That's kind of my really keen project right now. Okay. So, for those people that don't know what's synthetic data generated can you explain what about it?

Speaker 2: 15:05

Sure, sure, A lot of us do things like data masking or obfuscation, where we are replacing the production data which is real data, so we can say you know, your name is James, you know Liebman, but we replace that with John Doe.

Speaker 1: 15:18

Oh, it's like a test data.

Speaker 2: 15:20

Yeah, so we create that test data, but if you need it at scale, that takes a lot of work To do it with referential integrity and have it go. This reference key in this table is the same as this reference key here, so you can do joins and true development on something that looks like your production system but doesn't have the vulnerabilities is a big deal. The amount of and I did this to myself I had signed myself up for InfoSec breaches. I wanted to know what was going on. This will keep you up at night, no question about it. I mean all these breaches that are coming in every day.

Speaker 2: 15:56

Yesterday was 2.6 million records were hacked from a cryptocurrency. Day before it was healthcare. I mean, they're just every single day and I don't know if they're expecting customers users to just become numb to it, but I'm hoping that we can get more and more customers everywhere to start to mask their non-prod data, because that's where most vulnerabilities are. You find out somebody made a copy of a non-production database because, again, that's what's feeding AI and they're leaving it in their car and the laptop gets stolen, or they're sending it to a secondary vendor and that gets breached. So, to be able to do this synthetic data generation and then be able to use that for development and test data management. That makes a huge difference. Huge difference so that we're able to do this just on a laptop too. You don't need all this juice, because that's the whole thing with data masking is usually you have to have a dedicated server doing it, a lot of power, to be able to do that. So machine learning helps create something to do this that doesn't need a lot of power.

Speaker 1: 17:03

Okay, so I guess, in order to do genetic data generation, you need to do machine learning, and do you need to learn Python?

Speaker 2: 17:12

It definitely has a lot of python behind it definitely okay, okay.

Speaker 1: 17:15

So yeah, because the machine learning is predict what data is it? Because kinetic data is hiding test data behind real data. So, yes, you usually that see the interface will see the test data behind it will be like the real data.

Speaker 2: 17:30

No. So this isn't like dynamic data management views, which you can see in like SQL server, where the data is still present. You are completely replacing this so that your data isn't vulnerable at all. Dynamic data masking is still have a vulnerability to it and customers could have their data you know breach. You don't want that. So this is for full test data management. It is for a full replacement of that data. The production data no longer exists. This is for use for dev, for test, for QA. This is really powerful for a lot of developers who say I need something that looks like production, but I don't really want to have any access to the production data where I am accountable for this breach.

Speaker 1: 18:14

Okay. So, speaking of production data, are you using any AI in production workload for your clients, or are you just using PLC at the moment?

Speaker 2: 18:23

I do have a customer that's doing this. They are in healthcare and I love them. I like to say that they're really good at doing healthcare right. But as we were designing all this, we had to actually redesign it, because we were talking about you do development for AI a little differently than you do with standard multi-tier systems or more archaic client server systems, and that they were training the model all the way up through production, and they did use this. It's actually the same company that I'm working for.

Speaker 2: 18:51

It was their data masking that they were using, and they ended up poisoning their data model, because when we're doing AI development, we have to remember this that AI is always learning, and so when AI got to production, it said where's John Doe? I know John Doe, and they had to come back and go. Well, john Doe doesn't exist. That was just in our fake data, our test data we gave you. Well, I knew John Doe existed, though AI learns from where it is, so you have to almost scrub and start over in each iteration to make sure that you don't poison your data models once you get to production. So that was very interesting in how we had to design our development from beginning to end and make sure that we weren't poisoning our data models with test data okay, yeah, that's, that's great.

Speaker 1: 19:39

So are you, for example, are you doing anything? Oh, we're gonna say how can other people get started in ai? Do they need to learn any technical skills? Because I don't know, moment, mic Microsoft is pushing anyone into like into the playground to get a feel of it. Yeah, like whether it's AF Foundry. Do their people know that Python skill? Yeah, Python.

Speaker 2: 20:02

Python skills are incredibly powerful to have a good thing to have Python. I came from a bash world. Python's really easy because, of course, the syntax is simple. It makes it I call it the gateway language. It's very easy to pick up. Start moving forward with Python If you've got JavaScript skills.

Speaker 2: 20:19

Ruby's not really seen that much, but people are still talking about it and as well are those are your data scientists, and that I'm surprised how many people have data science degrees and are trying to get into it. But Microsoft gives a leg up. I feel that most don't. Microsoft Learn gives that capability of people can get very comfortable with things like Fabric. That allows you, if you come from, let's say, a background where you are more of a relational person or traditional development, that you're able to step from that relational world into Fabric and more your workspaces and actually working with you know parquet files and everything, and then bring in services that are low code, no code, and pull those in and be able to create chatbots and get more familiar with them before you go in deeper with AI, and I think that's phenomenal that they offer that. Most of the cloud vendors are starting to get there, but they're not as far as Microsoft is not even close to it.

Speaker 1: 21:16

Okay, so this year is all about agents. Just because we can do something doesn't mean that we should.

Speaker 2: 21:40

We need to look and say does this make sense? If you have a product that somebody comes in and says this you know right process, comes back and tells me my top five users at any given time in my system. If you're, let's say, a retail company, that's great Then to create an AI agent and to say, when you return those top five customers in my system now provide them with a code that will give them a discount, and you create the development code that goes back into the system and utilizes that and builds it and then gives them that discount it. That makes sense, but not everybody's going to look at it that way. I think we're going to have a lot more agents that don't provide a lot of value for customers.

Speaker 2: 22:25

I've got one AI tool right now that's coming up on my Apple phone that I would do anything to get rid of right now. All it does is get in my way. It keeps me from doing the things that I want to do with that application and that happens. That happens, I think, with every company, every cloud vendor. Somebody has created something with AI that just didn't need to exist. And you should always ask yourself just because you can do something. Should you be doing it? Somebody needs to kind of bring down that gateway and start thinking about AI.

Speaker 1: 22:56

I think they need to know whether it adds any value to the customer and have impact before they do it. Otherwise, we would just do the AI solution. So it's just made productive because they like it.

Speaker 2: 23:11

Yeah, and they checked off a box. They said there we go. I met my goal for the year. I have an AI agent.

Speaker 1: 23:17

Okay, so in a world of AI, ai is all about making people productive, and it can be known as the assistant from Microsoft. So let's dive into the privacy side, because it's very important. So how can we protect AI and stuff through modules as well?

Speaker 2: 23:37

Knowing what AI you're using in those tools. If you're using with co-pilot and using a private AI LLM, that you are ensuring that you're protecting that data is extremely crucial. Again, ai learns from the data that's there. If you are not sure about what you're doing, you need to bring in somebody who does. I am aware of some companies that one in the recent past that took lease data and opened it up to open AI public AI and you think about how much PII is in a lease document. It's a lot. That's really scary. They didn't understand that and all they were saying was let's see what AI says Shouldn't be doing that.

Speaker 2: 24:17

So most organizations should have a policy of saying what AI is allowed, that it goes through proper channels and is checked, and what can be provided to it. Some folks would say that that hinders innovation. At the same time, as I was looking into what AI products were produced in the last year, those that had strict AI policies many times had better quality. I saw that very clearly from the EU. Yeah, there was great stuff coming out of the US too, but there's a lot of extra stuff that I didn't feel we needed. That may have been kind of hampered, you know, and we wouldn't have had to kind of wade through it if we had had those kind of policies too.

Speaker 1: 25:02

Okay, yeah. So sometimes I get people like one important thing that you can do is like do the basic stuff like rather than protecting your module in, like using RBAC in Azure itself to protect the module itself and like do the least privilege itself for the users and things. Very much so, yeah, so you can also use things like content safety evaluation to protect your content safety.

Speaker 2: 25:30

It's just security elements of it, so keeping your data secure, like you didn't get misused or so an end point, um, especially if, if you're looking at data states that are outside and you're bringing them in let's say, you're going to do fabric and then bring in AI and that Make sure you're using your endpoints, that you can actually protect those, your firewalls that are there as well. There are so many different layers. If you can use a security layer that's available in a service, in infrastructure, utilize it. The more walls, the more protection methods that you can set up. It's like almost saying you know well, I'm not going to lock my car. Somebody's going to try to break in and they might drill out the door. Lock no, lock it. Make it as difficult as possible for them to get into it. Yeah, I would recommend that.

Speaker 1: 26:15

Okay, brilliant. So, as this episode is coming to an end, we always want to learn about our guests, so are you going to any events? Are you talking about AI? Are you going to any?

Speaker 2: 26:28

But it's probably not events that most of them know. I will be speaking at the DevOps Summit PowerShell DevOps Summit on August 8th. I will be doing the keynote at the Oracle Analytics and Data Summit. I will also be doing the keynote at Kscope, which is an Oracle event. What else I'm speaking at? Navigate 2025. That's also an Oracle event. I have an odd mix of PowerShell, oracle and Microsoft conferences coming up. Okay, and you talk AI in the PowerShell, oracle and Microsoft conferences.

Speaker 1: 27:00

Okay, and you talk AI in the PowerShell conference.

Speaker 2: 27:04

I will be yeah. Automation and AI yeah.

Speaker 1: 27:09

Nice. So do you have any hobbies aside from AI, your passion? I am a.

Speaker 2: 27:17

Lego freak. Yeah, there's a lot of Legos, a lot of Legos behind me. I am known for my Legos set and friends send them to me in the Microsoft community, which I greatly appreciate it. It keeps me busy. We have Lego characters. There's even a Lego me here. Yeah, we even have a Lego Kellen. My peers are here toove, jones, grant, fritchie, ryan, booze they all have their own lego characters that they put out at red gate every year nice, you'll probably run out to space suit.

Speaker 1: 27:50

You need more space I'm out of.

Speaker 2: 27:52

The shelves are full.

Speaker 1: 27:53

It's, it's bad okay, yeah, so do you have any last minute advice for people that want to go into a better organization, not yet dive into AI?

Speaker 2: 28:05

Take up the opportunity. Don't think about it too much If somebody offers you an opportunity to be part of an AI project. For me with the London Zoological Society, I was actually doing benchmarking on PostgresQL. I was that's what I was actually there to do and they asked me if I would like to work on the project. I ended up becoming part of the project and because I'm willing to do anything, I'm like oh that sounds interesting, I'll try that. It ended up creating an AI opportunity for me. That's been on every single project that I've been on. I am allowed to do projects outside of my primary job and that has just really paid off. I would say, anytime that you can volunteer, especially in the Microsoft community, that if you get the chance, learn Fabric. Fabric definitely has this gateway excuse me, this gateway kind of opportunity to other projects other technologies gateway kind of opportunity to other projects, other technologies. It is more analytics, absolutely, but it needs access to all these services, including just about everything in the machine learning and AI side.

Speaker 1: 29:08

Okay, so it's probably a great way to get into AI by learning data like Fabric. Yes, okay, brilliant, thanks. Yep. So thanks for coming on this episode, kelly. It's a pleasure for you sharing your story, sharing your story about AI and your journey, yep. So thanks for coming on this episode, kelly. It's a pleasure for you sharing your story, sharing your story about AI and your journey, yeah, so thanks Bye, thank you.