hans hoink animal science veterinarian arco laboratories non-coder

Our next speaker is Dr. Hans Hoink. He brings more than 25 years of experience in animal science as both an industry innovator and practicing veterinarian. I'm also an animal person, so if we've got that in common, love animals. Really excited to hear his presentation today. He has built his career at the intersection of science, leadership, and technology with a strong focus on applying AI and business systems to streamline development and operations. In today's session, he will share a first-hand perspective on using AI as a non-coder to build custom automation tools, including key lessons learned and how standardized processes enable scalable solutions. Please join me in welcoming Hans. Thank you for the introduction. I appreciate it. So I'm the managing partner of Arco Laboratories, part of the ownership structure, and also I act as the director of research.

small company purchase orders perishable product shipping deadline paul gormley

So we're not a huge company. We're a small company becoming a medium-sized company. That's the context of where I'm coming from in terms of how we're using AI. So the last presentation obviously was much more technical than what we're doing. But in terms of what we are doing, we're staying very practical. So if you're looking for something that can be done by a small team, this is probably a good presentation for you. I'll tell this story kind of as a narrative. Last year I went to, came to this conference and didn't really know anything about AI whatsoever. learned quite a bit here. And then after that, after the meeting, I went up to talk to Paul Gormley and just said, hey, you know, there's something that we'd like to look at. We get POs orders in basically every day, purchase orders, and we get quite a few. From when we started till now, we're shipping a lot more product.

order accuracy customer service staff expertise order verification competitive advantage

And so it can get to be overwhelming a little bit for our staff in terms of processing all those orders, which we have perishable product that's to ship out by the time UPS hits our door at noon. So we've got to get all that stuff ready to go back in shipping. Some Mondays is crazy. And so there's that, but there's also processing all those orders and processing them accurately. That's the main point, right? You don't want to send your product to the wrong customer. That's not a good look. So we're really focused on that. We do a great job. We've got double, you know, double check. But we've got these people up front that are just so good at this. They're really, really good at this. I mean, they'll get an order from a customer and our people up front will be like, that person doesn't usually order that. Maybe I should call them and make sure that that's what they meant to order. And a lot of times they're right. Like they send in the wrong order. Our customers love us because of that. Okay, that's something that we do that a big company not going to do.

cirrus automation project ai implementation architecture design diy approach

They're just going to send you the order. They're going to blame it on you. They're not going to refund it, you know, that sort of thing. So it's kind of our ace in the hole. Now, let's say somebody, you know, this is always, I don't want to get morbid, but you know, just the old adage is what if somebody gets hit by a truck and we lose that, you know, the redundancy of it. And so what we wanted to look at was could we model that into some sort of automated process using AI. And so I talked to Paul Gormley And initially he just said, hey, can you guys do this at Cirrus for us? And he said, yeah, we can do that. And this was, so this is about this time last year. And so we kind of went through the summer and the fall and he called me back. He said, hey, you know, I got a lot going on. I don't know if, he said, you can do this. He goes, you guys can do this. You just get an AI and ask him how to do it. I'll write you up the architecture for it. You dump it in AI and it tells you how, it'll tell you how to do that. And so that's what he did.

google gemini po reader quickbooks development team new year's eve

And I'm a Gemini guy, like I've drank the Kool-Aid. So that's what we used. And it was amazingly smooth. I had a team of myself and two other people. We were working on this. In the winter, we didn't get a whole lot done until New Year's Eve day. And that's one of those days that not much is going on, at most businesses. And a lot of people are, the people that are there are the people that haven't taken their vacation, or to have taken all their vacation. Those are the people that are there, but I made sure we had the team there. And so we wanted to do with this PO reader, I call it, or purchase order reader, was to be able to bring these purchase orders in that come by e-mail, read it, be very accurate about putting it into the system. So taking that and moving it to QuickBooks, which is our accounting software. And I learned a lot of lessons along the way. So this story has a twist in it, okay? It's not as juicy as Dateline, but it's got a twist in it. So anyway, we're going along here.

python vs code tesseract ocr optical character recognition technical architecture

We've got this, we're, we've got this plan. What we want to do, we've got the architecture from Cirrus, so we feel like we're going in the right direction. We dump all that into AI. And essentially how it was built was we followed this roadmap. Does this have a pointer on it? No. And so we use Python, which is essentially the engine of the car. We use VS Code, which is our dashboard, which shows you everything that you're doing with that code. And we use something called Tesseract OCR, which is a, I think it's an open source optical recognition. So when we get these POs in, sometimes they're scanned, sometimes they're digitized, but the machine has to be able to recognize them no matter what kind of format they come in. So that Tesseract was actually that part of the program that looked at it, said, is this scanned or is this digitized? Okay, I can read it either way. It's going to take me a little longer if it's scanned, but it can do it. So that's how it was built.

development sprint security certificate quickbooks integration api security non-coder mistakes

So New Year's Day, the 10 and a half hour sprint. So we started, I basically got everybody together. I said, let's see if we can do this in one day. Let's see if we can do it in one day. Okay. So we sat down. We worked. We did not take lunch. We just kept going. And I was like, we're going to have fun tonight. We're going to party. It's New Year's Eve. So we kept going. We kept going. We actually got it done to the point where we could get the PO reader to read the PO and put it into a format that would go into QuickBooks. And that's where we hit a wall. Because as I found out, as we found out, all of this software has security around it, right? So I'm not a coder. So if I would have been a coder or a designer, I would have known this from the beginning, right? So this was a lesson we learned, but sometimes these are the best lessons. because this is how you, I'm not going to forget that one, because we put a lot of work in.

certificate revoked security error star wars analogy jedi mind trick it support

But we got this, it's a certificate revoked, and we couldn't do it. I talked to our IT people, they couldn't figure it out. And so it's New Year's Eve, the hell with it, let's get out of here, let's go have fun. So we did, but this is kind of, let's see if I can, oh good, the sound does work. This is kind of what I was thinking later that night, like as I was imbibing and having fun on New Year's Eve. About 3 or four seasons? They're up for sale if you want them. Let me see your identification. You don't need to see your identification. Obi-Wan Kenobi, the Jedi mind trick? These aren't the droids you're looking for. These aren't the droids we're looking for. He can go about his business. You can go about your business. Move along. Move along. So that's what I needed. That's what I needed was Obi-Wan Kenobi to help me get through the security wall, right? Some kind of magic. Here's what was actually going on.

python expiration software updates security certificate version compatibility moving target

So another hard lesson learned. Software is not static. It's fluid. That's why we're constantly updating our computers, right? Well, this is one of the reasons. Python is a moving target. So this was the This was the version of Python I was using, which Gemini told me to use. And you can see it's hit its expiration date. Basically, its security certificate is revoked. So this is something that's going to happen. If we set up this software, this is going to happen over and over again. We're going to be dependent on it, and then it's going to break down. We're not going to know how to fix it. So at this point, I talked to Paul again. And he just said, hey, this is the point where you need to talk to a professional, a designer to think about like how to deal with this. Okay. So lessons learned. Again, talking to our upfront people. Is this DIY project really worth like corrupting all of our financial data?

risk assessment financial data standardization ai agent professional design

No. The stuff we were getting into is not very standardized. So it was going to be difficult for that to read that on every single PO we get and put it in correctly, maybe without an AI agent overseeing it, having a bunch of things that are set up. And it's a high risk, right? This isn't for a tinkerer. This is for a designer to do. So we did learn a whole bunch of stuff, though, in doing this, and this is the pivot, kind of the shift of the story, about how to do this, about how to use Gemini to code essentially. And what we were doing, which was crazy now when I look back on it, was we were letting Gemini coding partner code, you know, you whatever, cut and paste it and throw it in and then run the thing and see if it works. If it didn't work, then you go iteration, iteration, right? So the pivot was to go to this. There's another project that we were working on called a Google Sheets management system.

google sheets equipment validation regulatory compliance automated reminders legal attestation

And this was a system where we're in a very highly regulated industry. So we've got to validate all this equipment every year, sometimes by month, or sorry, by yearly, sometimes yearly, sometimes every six months, sometimes every quarter. But these are things that we do, and they're technical, but people forget how to do them because they haven't done it for a year, right? So there's SOPs on how to do that, but people forget about doing it. And we're very dependent on one person that knows exactly this schedule, and has it written out. But we don't have an automated process for reminding people and then having a feedback loop for those people to put in that they attest that they did it. Because you have, it's a legal record. You have to attest that you did it. And so what we did was We used a lot of this information that we learned on our kind of our failure with our PO reader to design this Google Sheets management system.

google ecosystem google forms gmail integrated tools sustainable design

And essentially what it does is it uses Google Forms, Gmail, and Google Sheets all together, essentially, to send out reminders. It's really kind of simple, actually. I mean, it's not as simple as I make it sound, but It uses an ecosystem that's already there. Does that make sense? Before, I was trying to create my own ecosystem that needed updated, that needed to do all that stuff. Now my designer is Google, and they're pretty damn good at it, right? So we got to thinking, okay, this is a better way to go. This is something that will last, you know, because we've got the kind of the built-in ecosystem already. And then the big breakthrough from this was, as I was saying, we were copying and pasting code. As we got close to the end of this PO project, we discovered this thing called Cursor. And Cursor is a heck of a tool. I don't know if anybody's familiar with it, but it is basically VS code.

cursor ide anthropic openai llm integration clasp tool

It's your dashboard. But it incorporates anthropic, OpenAI, you can incorporate all the LLMs into it, and it will pick the best one to write that code. And it's really good. It's really good. So what I had before was I had this spreadsheet that had a whole bunch of... Gemini was telling me, well, put this formula here in this spreadsheet, in this cell, but we'll also put in some coding, like app script, it's called, on Google Sheets. So it was this mishmash of formulas and code, and the more things we tried to put in it, the more it failed, basically, because it was becoming more and more complex. Well, I put in cursor and then there's this class tool that essentially pushes the software, pushes the code into Google Sheets. So I didn't have to cut and paste anymore, which was really nice. But what it did was I told it, literally, I just said, this thing is full of, you know, this thing's a mess.

cursor refactoring code transformation spreadsheet formulas google apps script automated coding

Like this thing has all these formulas in it. tied with code and it doesn't work. And it went in and replaced all of those formulas with code. So it's now it's completely coded. It did it in 10 minutes. 10 minutes. This was like weeks of work. And it just made it better in 10 minutes. It was unbelievable. So I'm A cursor believer. But now we're using this tool as a very effective way. And we're depending on this tool now because it's extremely effective. in reminding people there's a feedback loop. It is a legal record because there's an attestation that goes along with it that says you attest that you did it, and it's linked to your, basically your digital signature. And so we can show a regulator, here it is. So it's a really, really nice tool. This is what the dashboard looks like. I mean, it's very basic, but it tells you, oh shoot, sorry.

dashboard task tracking regulatory audit digital signature documentation requirement

Like up here, it'll tell you what tasks are pending, what's upcoming, and most importantly, what hasn't been done. And in our industry, if it's not written down, it wasn't done. That's how we look at it in terms of highly regulated industries as we are, because we have to be able to show somebody that we did it. So the lessons learned from the PO reader, basically you can't automate a mess. Like it was, especially with just code. There was just too much craziness going on in terms of the difference of all the POs we get in. And so one thing we realized is we don't, I didn't realize this, but we don't have individual numbers for each product. People were just sending in the names of the products. I didn't realize this, but it brought it to light. It was like, we got to get some things in place here before we start automating, before we ever do that.

product numbers standardization distributor training data structure automation prerequisites

So end of 2026 is our goal for getting unique numbers, training all our distributors on how to order our products so that we can do that. But without that, this isn't going to work. The PO reader isn't going to work. So that's a longer term project. So essentially that was with the PO reader. And then in my industry, we always have practice tips at our CE, for veterinarians, practice tips. These are my practice tips for using AI. The screenshot is your best friend. So if you're looking at something and you and the LLM aren't, you aren't talking the same language or it's trying to tell you one thing and you're like, no, that's not what I'm telling you. Just take a screenshot and throw it in there. Because what happens a lot of times is you'll be looking at something on that screen and you think that's the problem. You take a shot of the entire screen and Gemini will be like, no, look over here, like up here on the left.

screenshot debugging llm communication ecosystem approach google integration non-coder advice

That's your problem. You don't have that. So it's extremely intuitive in terms of solving those problems. If you take that screenshot, you say, here's what I'm looking at. Now you look at it too. That was really helpful. Accept your role. I'm not a coder, and I shouldn't be messing with data that could sink our whole business. Like now we can't take orders, right? Because I just messed up the, whatever, the QuickBooks system. And I did get a lot of guff about that from our upfront. People are like, do not do that. We'll quit. Thinking ecosystems. So if you want to keep things simple, you know, my first example is how not to do it, kind of. The second example, we got a little smarter because we knew this was more robust. I guess they say that a lot in the computer world, but it was more robust because Google has the ecosystem set up. And it also has Gemini, which is so now we can possibly start to use Gemini to weave it into our management system as well, because all of that is already integrated.

risk management hire professionals critical data system downtime beta testing

Don't climb walls alone. So again, on the PO system, don't get stubborn because you can get yourself a lot of trouble and then protect the core. So critical, critical data is critical for a reason. And so you don't want to start doing DIY projects on stuff that you got to think about what you're working on and then think about what would happen if you ruined it and what repercussions would have. I mean, you can back things up, but when we're getting tons of orders every 5 minutes on Monday morning, that's not the time to be down, even if you have everything backed up, because that can be a mess. So you always have to think about that. You have to be in beta mode. And when you get to that point, it's good to hire somebody, hire a professional. So that's my presentation. Thank you.