Maktelier Founder Paul Breuler is an interesting man, to say the least. What he’s doing in the realm of AI is even more eye-catching than his background, perspective, and approach to life— all of which have led him to dedicate 16 hours a day to his recent efforts with the fully automated warehouse solution Pallet Shuttle Automation, avoid almost all forms of television, and still find time for his passions like woodworking while raising a young family with his wife, Lindsay Breuler, who, as managing director of the gener8tor North Dakota Investment Accelerator, also lives a vigorous life.
“How do you become a Thomas Edison or a Nikola Tesla? They started young. They didn’t have TV. They read a lot. They learned a lot. They got involved,” Paul said. “They never stopped. They didn’t vegetate in front of a television like we do today. I try to live by that. I try to be moving and be with my kids—I try to be doing things.
With that approach to life, Paul has already been able to accomplish a lot. After overcoming early struggles in life that included experiencing homelessness for a time as a teenager, Paul managed to rise through the ranks at Microsoft as a senior software engineer building the Application Lifecycle Management (ALM) for Power Platform workshop and services, a multimillion-dollar business within Microsoft.
About Pallet Shuttle Automation
Pallet Shuttle Automation, founded in 2008, is an advanced storage and retrieval solution designed to optimize the handling of goods within warehouses and storage facilities. It’s a part of automated warehousing systems that aim to increase efficiency, reduce operational costs, and maximize storage density. Paul Breuler and his team are working to make the solution even more powerful through the deployment of AI.
Toward the end of his work with Microsoft, though he didn’t know he was nearing it at the time, he was approached by a friend and began working with Pallet Shuttle Automation.
“My friend called me because the co-founder of Pallet Shuttle had passed away and they needed some help,” Paul said. “They contacted me last December, and they said they needed it working in two months so I was like, ‘Okay, let’s do this.’ I started off just moonlighting as I was working at Microsoft because I figured I’d just help them out a little bit. Originally, I told them I’d give them my afternoons and evenings. There were a lot of late nights, but the algorithms are well-documented and publicly available. So, it’s not like it’s anything fancy or novel in terms of getting things to move around. It’s very well known. Those algorithms have been around since the 1960s.”
What was one more project for a man who had been tinkering his whole life? Paul grew up pulling apart computers and putting them back together, taking apart RC cars and piecing them back together, and “doing things he shouldn’t be doing with computers at school.”
“[The school] didn’t have the security that they do today,” Paul said. “We were able to get on the terminal and spam the whole school’s computers with all kinds of random messages. It was just fun stuff but it was that hacker mentality and trying to get things to work in an automated fashion.”
There was even a time when Paul was wrongly accused by his father of sabotaging his own van so he could drive his father’s car. Because of this, his dad stood over his shoulder and made him fix everything himself. “I didn’t like getting yelled at, but I loved the work. I loved figuring something out that I’d never done before.”
The Next Thing to Figure Out
Paul only moonlighted with Pallet Shuttle for about a month and a half… because he ended up in the hospital.
“My wife decided to move back to Fargo just that year prior in September and we wrapped up a five-year home renovation. She told me about 60 days before she wanted to move that we needed to be out and done. I somehow condensed two years of work into about 30 days. I worked all day, every day. At some point, I got hurt because the hours were just too long. I fell off some scaffolding, tore some ligaments, and got seriously injured. The injury, the stress, and the move to Fargo all just killed me. So, I was in and out of the hospital for two to three months. It was pretty wicked, but it gave me distance between Microsoft and Pallet Shuttle and gave me time for reflection.”
"How do you become a Thomas Edison or a Nikola Tesla? They started young, they didn't have tv. They read a lot. They learned a lot. They got involved, " Paul said. "They never stopped. They didn't vegetate in front of a television like we do today. I try to live by that. I try to be moving and be with my kids - I try to be doing things."
That time for reflection didn’t immediately facilitate a career change for Paul, who then returned to Microsoft for a while, but according to him he “just wasn’t there mentally.”
When you ask him about his time at Microsoft, you can tell that he has true passion for the work he did over his seven years with the tech giant and that he admired the people he worked with. During Paul’s time at Microsoft, he wrote and authored materials with another software engineer named Melody Universe that helped companies and governments deploy code in a drastically shortened amount of time.
“Melody and I authored the ALM for the Power Platform workshop. We created multiple services and authored a good bit of tooling, such as libraries, templates, and the core logic for Power Apps Command Line Interface (PAC CLI) that allows you to write code for the CLI once and deploy to Azure DevOps and GitHub from the same code base,” Paul said. “We created and scaled the Workshop, services, and consulting business with over 100 accredited engineers globally to deliver, evangelize, and support customers’ ALM needs. A large group of amazing people worked to scale the ALM solution at large along with us—I want to make sure that people know this as well.”
We spent two years with the Government of Canada and when they went to production, I took their entire tech staff out to a cafe and bought them all coffees, lattes, and some snacks. We brought a laptop and one cell phone and we did deployment for the entire Government of Canada Economic Development Committee from a cafe in 45 minutes with a cell phone,” Paul said. “The ability to unburden people is an amazing feeling.”
However, upon returning to work, Paul’s ALM team had been decimated by rounds of layoffs within Microsoft and he found himself disillusioned and amid a professional identity crisis of sorts. He had lost one of the things that gave him purpose. A thing he had found stability in after a lifetime, or as he says, “many lifetimes” of instability
“It was the culmination of many small moments that brought me here because my life story is just insane. I’ve lived many more lifetimes than a person my age should have ever lived,” Paul said. “I worked a third shift doing baking. I’ve gone to school for cooking, and architecture, and eventually got into computer science. However, I started a computer science-type business doing small business networks in my teens.”
When I met my now wife, Lindsay, I was 15 or 16 and I was homeless previous to meeting her and had just gotten back with some family who were helping me figure things out. I wasn’t doing the things I was supposed to be doing and she said to me, ‘I don’t date stupid people.’ Prior to that I would literally show up to school, take a test, and I’d walk out the door because I was hungry and I didn’t have a house. So my first goal was to eat. I would skip school and I would go to somewhere like McDonald’s and eat. I wasn’t applying myself because I was looking at things day to day to get through the turmoil and to get through my life. But one day, Lindsay said to me, ‘Take a step back. Where are you going to be in five years? Because you won’t be with me if you keep doing this shit.’ After that, my grades went straight to A’s and I skipped from algebra to geometry to calculus. I was taking anatomy and physiology honors and doing actual dissections on cadavers in high school. I went from an idiot to a very high level overnight because I had somebody who believed in me and who gave me that backing. So, if I were to say there was a pivotal moment, it was meeting my wife, hands down.”
What is a Distributed System?
Distributed systems involve multiple software components or applications that work together on different machines or environments to achieve a common goal. This architecture can enhance the system’s scalability, reliability, and fault tolerance.
Paul says Lindsay has continued to push him to be better through the years. “I wouldn’t have started the first business if it wasn’t for her. I wouldn’t have gotten good grades in school if it wasn’t for her. I wouldn’t have scored 34 on my ACT if it wasn’t for her—my teachers thought I cheated. There was somebody in there who had a couple of brain cells to rub together, it just took someone to wipe off some of the grit and smack me a couple of times to get me there because living on the streets was everything you don’t want your kids to have to go through. I’ve been stabbed through the chest. I’ve been shot at. I’ve been lit on fire, all kinds of shit.”
When you overcome those things to get to where Paul has gotten, you are never one to sit still for long. In June of 2023, he left Microsoft and founded his own company, Maktelier, which would contract for Pallet Shuttle full-time.
“With Maketelier, I started developing a new algorithm based on all of the things that I had learned in the past,” Paul said. “I’ve figured out the guts of how to pass these systems that no other competitor (that I’m aware of) has today. I’ve done hundreds of pages of marked-up research and all the competitors that do these things all have the same problem. They all hire mechanical engineers, electrical engineers, and all these noncoder types—they don’t look at it as a computer science problem. Problems in warehousing logistics have been solved for 40-50 years, but the technology just hasn’t been applied because, just like they do in agriculture, in manufacturing, things move slowly. “They’re the staples of our society through which we are able to support ourselves, but for whatever reason, we don’t invest in them. So, by taking a step back and asking how I could look at it in a more simple manner, I was able to develop this algorithm over a few months, that now forms the basis of what Pallet Shuttle is operating on.”
After developing that algorithm, Paul sold it to Pallet Shuttle Automation and agreed to a Professional Service Agreement contract with Pallet Shuttle to work as its interim CTO through the end of June 2024 to help the company get to production.
“So, it took three months to build out the core of that algorithm,” Paul said. “Now it’s about building the rules that help support all the different use case scenarios. I want to be clear, we are not building a separate solution for each customer. That’s what they were doing in the past. That’s not supportable over 30 years. As a manufacturer, you’re not working with one-year deals. It’s 10-year, 20-year, 30-year deals. Our smallest project is 10 million pounds of steel. You’re not getting rid of that in 10 years, you’re not getting rid of that in 20 years, it’s going to be around for 30 years, which means your software has to run for 30 years, which means your AI has to run for 30 years.”
What is Iterative Learning?
Iterative learning in AI refers to a process where the learning algorithm iteratively improves its performance by repeatedly going through the data, learning from previous mistakes, and making adjustments to its understanding or model parameters. This concept is fundamental in many machine learning and deep learning methodologies, where iterative processes are used to optimize models to perform specific tasks, such as image recognition, natural language processing, or predictive modeling.
And this is where AI truly comes into play. By building the solution the way Paul is building it, Pallet Shuttle’s solution will follow a core set of rules which he sets. However, on top of that, individual companies that purchase the solution are able to apply their own sets of rules and business logic (FIFO, LIFO, etc.), and the AI will use iterative learning on top of that—according to Paul, that’s where the magic happens.
“Iterative learning is where the if/then decision tree comes into play, but it improves because the AI will learn over time which decisions are good and bad,” Paul said. “The iterative learning algorithm has been around for a long time. If there is a problem that exists today, it probably has already been solved in computer science. It may not have been commercialized or put into a product, but I guarantee you the solution already exists. It’s just a matter of how to apply that solution and how you map it onto a real-world problem.”
Recently, Pallet Shuttle took the next step in development, going from its base algorithm to a distributed system, which is crucial for having a warehouse full of robots work together.
“The hardest part of everything is not getting things to move—it’s getting things to move in coordination,” Paul said. “Imagine you have 20,000 Teslas in New York City where everything is a one-way street and you don’t have humans involved and traffic has to continuously move without anything ever stopping and you can’t have any collisions— that’s what we’re solving.”
The Pallet Shuttle solution is essentially a series of oneway hallways, which is dramatically more complicated than solving for an open-world problem like an autonomous vehicle because they can move a little to the right or a little to the left; whereas everything in the warehouse is limited to forward, backward, left, and right. “And you have to know, ahead of time, that the decision you’re making is reasonably good to keep the performance going,” Paul said.
Pallet Shuttle has essentially created a working solution for this already. It’s just a matter of testing and deploying. However, that does not mean that the company or Breuler are done innovating.
What is Multi-Agent AI?
Multi-agent AI refers to systems or models in artificial intelligence where multiple agents interact within an environment to achieve individual or collective goals. These agents can be software entities, robots, or any intelligent system capable of making decisions and taking actions based on their perceptions of the environment. The interaction among agents can be cooperative, competitive, or neutral, depending on the system’s objectives and the nature of the tasks.
The Next Horizon
“This is where things get hard,” Paul said. “To my knowledge, and based on Massachusetts Institute of Technology’s (MIT) latest publishing as of May of 2023, the realm of multiagent AI is largely unexplored and is kind of a horizon for the next 10 years. I’m not trying to solve all multi-agent pathing. I’m trying to solve it in one very small, very particular use case where we have one-way hallways and robots that can go front and back and left and right and that’s it. And it is still tremendously complicated.”
“Each robot can path on its own,” Paul said. “In a scenario like this, if you imagine the number of collisions on a graph, it starts real slow but as you go up in the number of robots— to like four or five or six of them—you get to a point where no robot can ever path because every move you make puts you through the path of another robot. Solving that part is another layer—it’s called hierarchical pathing. Every robot is like, ‘I want to do this path, I want to do this path, I want to do this path.’ And you have to figure out, based on the math, if they are all free to do what they want or if there is contention. If there is contention, you need to figure out how to solve that.”
To do that, according to Paul, you can apply a genetics algorithm to “to spawn all these little islands” on the map and make them battle it out to figure out who has the best decision. Then, all those islands die off and based on the decisions that win, you have a new evolution of islands on the map. “It’s essentially applying different weights and logics around the map [of the warehouse] to figure out if it makes these different decisions and what will result in the next best action. And that has to be done every second—it’s a constant thing that’s happening,” Paul said.
“The nice thing about our solution is that once a robot is in motion, it performs that entire motion as one movement, because it only goes forward, back, left, and right,” Paul said. “Say a robot needs to move to the left three positions, it’ll take time—it takes like two to three seconds. That gives my AI two to three seconds to munch on everything else that’s going on in that whole system.”
According to Paul, the opportunity to play a part in exploring these new horizons is what keeps him motivated and working until 3 a.m. some days. However, much like his AI, he is taking an iterative approach with Pallet Shuttle.
“I actually figure you don’t need this multi-agent pathing to get something into production,” Paul said. “You don’t need it because most customers have very simple requirements. You can zone things so the robots each have their own little world where they’re operating and then they never collide— that works for most cases. Before we get to multi-agent AI, we can do something naive in between—we can make simple decisions based on priority. Right now, if a robot needs to go forward and other robots in the way are in a lower priority, we have them get out of the way. I call it the GTFO solution. That’s literally what it is. You have to have shuttles to be able to GTFO in order to solve the rest of the problems.”
“You have to take this iterative approach because of the complexity of where we are heading,” Paul said. “If you can solve a business problem without going to the nth degree, do it. This isn’t academia. This is solving problems for real-world business use cases. But we’re going to continue to iterate.”
And as he and the Pallet Shuttle team continue to creep towards that nth degree, he’ll be well equipped to deploy the updates along the way “Melody Universe and I authored the Automated Lifecycle Management workshop and services for power platform (ALM),” Paul said. “We traveled the world teaching customers how to deploy software with the least stress possible. So who did I hire to help? The person who got nixed by Microsoft around the time that I decided to quit to be my ALM guru, if you will, Melody Universe. So Melody Universe and I have a solution she demoed to me last week, where we can deploy the entire solution from zero to 100, in less than two minutes. Melody called me up and said, ‘I got this to work and text does not convey how excited I am.’ She showed me the deployment and I’m like, ‘That is the sexiest ALM solution I’ve ever seen.’ This is the culmination of the seven years we spent building the ALM business I was talking about. So, not only do we have AI, but we have the ability to deploy that AI and a net new solution in minutes, and have everything running. Which also means if there’s ever a disaster, we can recover in minutes.”
Things are moving fast for Paul and Pallet Shuttle Automation as they continue to work on the cutting edge of robotic automation and according to Paul, he has already documented the solution. It’s just a matter of testing and deploying.
“I want to be able to put the robots in a building and have a door go in and a door go out and you don’t even have to know what’s going on inside,” Paul said. “All you have to know is that there are pallets coming out the other side.”
Follow along as Paul and Pallet Shuttle Automation work to make this happen.
Pallet Shuttle Automation
Paul Breuler
Linkedin | /in/paulbreule