Steven Zgaljic, Jahnel Group, CTO: Harnessing AI for Health Insights

Speaker A: Hey, before we jump into today's show, if you're a CTO looking to stay sharp in an AI powered world, join a select group of tech leaders at the 13th CTO Colloquium in Austin, Texas on April 17th. This is a two day event. It's packed with powerful sessions like the AI Adoption Ladder with Dan Martell Forecasting LLM Best Practices with Scientist.com CTO Chris Peterson and Building Agents with MCP from Protopia AI's Jennifer Swagenberg. You'll elevate your leadership, sharpen your strategy and connect with peers who truly get it. Head over to Sevenctos.com to learn more and register and I'll be there and hopefully I will see you. So let's get back to the show from 7ctos. My name is Etienne De Bruyne and you're listening to the CTO Podcast. Every week I spend time with fascinating people that enrich the lives of chief technology officers around the world. From perfecting the basics of building technology organizations to inspiring our minds into shaping our future. As always, the CTO Podcast is brought to you by seven CTOs helping CTOs become world class leaders. Let's go. Stephen Jigelik. After. After all these years. Am I saying it right?

Speaker B: It's perfect, it's organic, it's natural is how you're supposed to say it, but you just plow right through.

Speaker A: Uh, remind me, do you speak the language?

Speaker B: I don't. It's a big disappointment.

Speaker A: I know.

Speaker B: It's very good.

Speaker A: Well, well, Stephen is, uh, CTO at, uh, Janelle Group. You guys build awesome things for massive companies.

Speaker B: Yes.

Speaker A: Hey, you, uh, and I were on Slack this morning and you shared a pretty touching story. I don't know how comfortable you are with sharing this, but I feel like awareness. And there might be some people out there who might need to hear how you're going about this. So why don't you share a little bit and um, if this is gets too intimate, we'll just cut it.

Speaker B: Okay, no problem. So I have two kids. I have a seven year old girl and a five year old boy. Wonderful kids. And recently this year my 7 year old has been experiencing, I would just call it a mysterious illness. You know, I think everybody has experienced something with the latest flu season and all the viruses going around. So it's no surprise to many families out there your kids are experiencing a lot of stuff. So she had an infection earlier this year and we got through that, but these symptoms sort of pestered and they lingered and they kind of Came and went. And as it's been going on, it's been escalating a little bit more and more. And basically, over the past couple of weeks, my daughter's really been missing school. And she's, I want to say, debilitating, but it's to the point where it's really like, you know, upsetting her daily life. She can't go about her day, she's not going to school. So, um, you know, me and my wife are going to the doctors a lot and trying to understand what is really causing this right now, because she's doing the typical tests with the doctors, blood work and all that, and we can't really pin this down. And really where I'm at with the doctors is, well, you gotta really take notes and just kind of monitor throughout the day and what is she doing, what are her daily activities, what is she eating, when is her stomach getting upset? And trying to understand the behavioral aspects and the interdependencies throughout the day. So really, the doctors have us tracking her every move, and I think everybody can understand how difficult that is. So, like most folks who've done this, you know, we started off with the pen and paper. And you, uh, wake up in the morning. She woke up at this time. This is what she ate. This is when she went to the bathroom. And kind of tracking all of our symptoms throughout the day. And you move around and it just gets really difficult to track that paper, Especially with two people, me and my wife. So we've graduated out of the pen and paper into the digital world. So we made a Google spreadsheet and a Google document that gets messy to its own, right? A little bit more structured with Google sheet. And, uh, it just struck me actually, the other day when I was at the doctor, I had a really nice polished Google sheet because I'm ocd and I keep things as ordered as possible. And she's sort of asking us, you know, verbally, like, well, tell me what's going on. And like you do when you go to the doctor. I was explaining it to her. I'm like, you know what? Let me just show you the sheet. So I actually, like, showed her my phone. And she was like, oh, wow, this is amazing. Exactly what I want. I had a timeline ordered, categorized, and this was me doing this myself with the Google Doc. And she was blown away. And the doctor came in and they were like, you know, nobody has this information, this organized. So I went home that night. I was like, that's, you know, everyone's. I'm a Tech guy. Everyone's playing with AI these days, myself included. And I'm like, this is exactly, I think what AI can be used for. And it could produce a report like this and it can help me keep track of that. So I just went into the lab that night, you know, and I was able to build a very quick little solution with N8N super base and slack. And my family uses Slack to uh, kind of communicate the family ordeals, coordinate school activities, you know, what are we eating for dinner, those sort of things in there.

Speaker A: That's actually interesting. That was the first thing that. Well, the first thing that struck me was you being willing to share the story and. But uh, then I was looking at your little flowchart, which I want to get into. But I was like, oh, this family uses Slack as a family. We have a group text and boy, it is a nightmare to get my kids to read the group texts.

Speaker B: Yes.

Speaker A: Okay. So anyways, you guys adopted Slack, the paid voyager.

Speaker B: No, no, no, we're not paying for that. And my kids are not on there, only five and seven. So maybe that gets more interesting. So you're right.

Speaker A: So it's basically uh. Oh, like for the, the, the larger family as well.

Speaker B: So I've tried with my extent, my brother and all that, we had that at one point. I haven't cracked that code with them. This is really between just me and my wife and like the things within the home. So, you know, little channels that are dedicated to different things. I'm really the champion of that, to be honest with you. But she's partaking in that and she's entertaining that I think it's a powerful thing.

Speaker A: Perhaps he's so powerful because the, the number of times, the conflation of channels, conversations, like where did you post the grocery list? Wow. Anyways, that's something I'm going to try out. I wonder how many other people are doing that. So it's just the two of you, but you have, ah, the benefit of channels.

Speaker B: Okay, the benefit of channels and the benefit of automation. And now I guess with AI, the benefit of. You're right, it's not just me, but it's me, my wife and endless possibilities out there. Right? We could create all these agents, like

Speaker A: post something to the groceries channel and boom, it's ordered. It shows up at your door. Boom.

Speaker B: Yeah. What are we going to eat tonight? And it's no longer just you and your wife, but it's, it's participating and maybe you to order out. Maybe the A. I haven't built that yet, but the AI could potentially order out for us. So it really opened my mind after I made this nurse app and the nurse understand this is a context of which child I'm speaking to, because the channel is dedicated to my children. And uh, yeah, as I'm talking, it is understanding, you know, is this a symptom? Is this a food thing? Is this, uh, you know, a restroom activity?

Speaker A: So hang on, you went back and being the CTO type that you are and the entrepreneur that you are, you are like, okay, I'm going to create an app. Are you ready to elevate your team's performance and your own career as Ah, CTO? Introducing the 7ctos Growth Program. With 7ctos, you'll join a supportive community of tech leaders dedicated to helping you reach new heights. Our, uh, group coaching program led by experienced CTOs, focuses on the proven CTO levels framework. Assess your team processes and technology challenges with our easy to complete assessments. And then dive deep into monthly coaching calls where you'll get unstuck and gain insights on specific topics. Craft actionable plans tailored to your needs, guided by expert advice and playbooks designed to drive results. Plus gain access to our exclusive CTO community for ongoing support and collaboration. Sign up now and start your group coaching Adventure. For only $300 per month or $3,000 per year, your membership includes priority access to all seven CTOS events and activities. Don't miss out on the opportunity to join a global community of experienced CTOs, attend virtual events and participate in confidential group coaching sessions. Visit sevenctos.com growth and level up your leadership skills today. And you said you used N8 in, uh, Superbase, so. So walk me through. You're sitting at your desk like a. How did you decide on those tools? I think a lot of people are wondering. There's so many tools now. Like, is it just something that you guys have been exploring at Janelle or what?

Speaker B: Yes, definitely. I mean, these tools, Nan, specifically like the workflow automation. I've used make.com, great technology. It's great. But Nan has been something that we've been playing a lot with recently. So we have doing a lot with our recruiting pipeline and I've actually been building a lot of things in our sales pipelines in order to help automate the first, uh, point of contact, the first touch point to code. So I'm almost. I haven't come up with a beautiful word, but it's like clients to code. Like, how fast can I sort of expedite and enrich that experience with AI and having agents all throughout that. So we were deeply in that space and it was really just like, you know, new project. I was able to spin up that workflow and I was very familiar with that right now. So using that and yeah, the agent workflows, you know, open API and it's crafting those prompts. And I'll tell you, there's a lot of lessons learned in here. I have a lot that I've learned in there. The tool agents and how you're able to instruct your AI to take advantage of a tool. Mcps, the new thing lately, but this, I'm just using traditional tools. I've noticed you have to really uh, convince your AI to take advantage of tools. A lot of times it's aware of it, but it may not really like create an event. So I would say, hey, you know, you know, she woke up this morning and you know, she ate some food and had a stomach issue later. And the AI would tell me yes, it was created and I would go look and it actually wasn't created so I had to do something more interesting. And I had two AIs. So one AI is the first one that makes the event. And I have another one called the Valid Validation which basically goes back and looks at it says, I'm hearing that,

Speaker A: I'm hearing that more and more. It's like the um, there's a whole science around just the validation models, the error models. It's really interesting.

Speaker B: Absolutely.

Speaker A: Just quickly walk me through just literally how you built it. Just two minutes.

Speaker B: Yeah, two minutes.

Speaker A: So basically you went to N8N.

Speaker B: Yes, go to NAN, you create a new workflow, you have to configure accounts, you have to connect that into Slack, you have to connect that into, you know, Open API, have all those credentials there and then you basically create an agent and from that agent you have to define the first of all the tools that it's going to take advantage of. I started off with a Google spreadsheet to manage the backend there, but I've noticed some trickiness there. So I created tools and I connected with Supabase, which is basically a low friction postgres database. So the actual events are going to be stored inside of a postgres database and other feedback that I get from a doctor goes in there as well. So when you create that nan, you connect the tools associated to it and really the secret sauce comes down to the prompt that you give the agent and describing that prompt. It's quite extensive. And I basically used AI to iterate through that process to define it, I had my ideas and I kept iterating back to AI to get more and more refined system prompt there. And that was the uh, gist of it really, in trying.

Speaker A: Is the system prompt something that, that is sort of like a rag style, like a pre prompt that, that, that gets loaded before your actual recording of the symptom. Is that what you mean?

Speaker B: Yeah. So when you create an agent and you put it there, the agent you will define it with like a system prompt. And that basically means any message that comes into the system, it's going to load that message first.

Speaker A: Yes. Okay.

Speaker B: And then as messages, new slack messages come in, it's going to get appended afterwards. You have to have this concept of a memory there. So N8N facilitates very easily for you to create an agent and attach memory associated to it. So if I send my initial slack message, it will load the system message plus my first slack message. If I come back with my next slack message, it's going to reload the system message, but it will pull from memory of all the past events that I've messaged the system. And it's great because it allows you very easily to create these window buffer, window buffer memories, which is literally Nan tracking that for you. If you've done this yourself with LangChain or other technologies, you have to code it. It's much more difficult to create that memory system there. So Nan made it like very simple. Literally. Plus in memory. How many messages do I want the AI to remember? I put 10. So the AI will always remember the last 10 messages. But that's more than enough for the context that I have for what I'm dealing with.

Speaker A: And the, uh, context of 10 messages is important in the sense that if you drove to the doctor's office or you went to a, uh, McDonald's or you did something like, you can just have the context maintained of. I was at the restaurant. She started displaying this symptom and then just. And then ultimately is the end goal to be able to then produce this cleaned up spreadsheet to the doctors, or are you going to feed it to the AI and have the AI tell you what's going on?

Speaker B: So all throughout the tracking, I've actually instructed the AI to say, absolutely, first and foremost, take what I'm telling you, record the time that I'm telling you, derive the time of the symptom. So I may be telling you, hey, earlier today, around 9:15, she had a problem. So it would know I, you know, I informed at 3. But the symptom was at 9 and take track of the symptom. But there's another column in the database that say, give me your AI insights and tell me if you think this is a new sentence, is it reoccurring, is it linked to something else? So each event, AI has drawn insights against that. And I do have a plan later to come back in and ask it like more formally, please give me a summary of what the past week looked like, what the past month looked like, or something like, or I'm at the doctor's office, I'm at this particular doctor. Take insight of everything that I've given you. And what would a, uh, GI care about? What would infectious disease person care about trying to do that? So that's next.

Speaker A: But yeah, it does strike me how there's this promise of AI tools that, oh, it's quick and easy, but then if you start digging into, creating these prompts, connecting these tools, you, it's pretty sophisticated. Like you, there's a first principles thinking or a, shall we call it a logic engine that you need to have as a human in order to be able to extract this from the agents or from the AI.

Speaker B: Absolutely agree with you. That's actually what's on my mind this morning and I was thinking about, I was like, wow, this is uh, a whole new set of tooling, a whole new set of design patterns, architecture, thinking, system design around this. As a technical engineer, ctos we've been building with really structured code blocks, software technologies. This is more human oriented, right.

Speaker A: And it's actually more, I think it's more spec driven or document driven as Ryan Weiss would say. In our community, like Ryan Weiss is advocating the DDD stuff, but it's this document what you want, see the code, invest it back into the document so that in the end you kind of have this incredibly full spec document. Have you seen anything yet from the, from. Has the AI revealed anything special to you yet from the data that you've invested? Well, it's about your daughter at all or have you seen any early success?

Speaker B: It's a little early to tell. I made this recently so I'm back filling a lot of the data in there. But it definitely is showing me, you know, really interesting trends and certain times of the day where something may happen or you know, correlation between a particular type of food is there. But I think there's a lot of promise and there's a lot of hope and I'm um, I'm excited to continue using that. But yeah, it's doing what I'd want it to do so far.

Speaker A: I love that. And just quickly, are you using primarily text input? So it's just. It's through the slack as the event?

Speaker B: Yeah, 100%. This is just natural language. I'm naturally speaking, but this isn't like

Speaker A: leaving the thing and saying, hey, computer just left the doctor. And she had, you know, it's not

Speaker B: like speaking to it that I thought about it, but I'm just typing right now. That's the next step. Next step is just my daughter and the phone, and she's just gonna speak into it.

Speaker A: I love it, man. Is your wife using it yet?

Speaker B: She is. I put it into her hands and I'm like, okay, keep, you know, keep tracking. Yeah.

Speaker A: Well, I know as a parent, it's terrifying when these things happen to our kids, and so kudos for this path and I really appreciate you sharing this and I'll keep. I'll keep my eye on you and see how this goes.

Speaker B: I appreciate it, and I hope this inspires others or gives people other ideas for whatever they're faced with.

Speaker A: I love it. Okay, Steve. Take care, man.

Speaker B: Thanks.

Speaker A: That's the show. Check out ctopod.com. stay connected. We love hearing from CTOs. We love hearing from CEOs. Anybody who needs to get their CTO plugged in, check out sevenctos. Com. There are membership levels for everybody. So it's never too late to expand your network, nurture your relationships, and please, let's see each other soon, like next week. Cheers.