ACCEL-KS Profile: Freeman Constructs, Inc.
As you’re probably well aware, we were awarded a grant from the Kansas Department of Commerce to support early-stage founders in Kansas. With that funding, we designed the ACCEL-KS Proof-of-Concept Grant, a program that provides up to $25,000 to help Kansas founders take their projects from idea to commercialization. Our program includes wraparound services with ecosystem partner support from across the region. We received far more applications than we expected, and after a rigorous review process, we selected 20 grant recipients. Now, we’re going to share more about the projects they’re working on. We asked each founder the same set of questions, and over the coming weeks, we’ll publish their answers.
Up next? Aaron Freeman.
Company name? Freeman Constructs, Inc.
Product name: Tenfold AI Accelerator
What gave you the idea for your product or startup?
A friend of mine and I were experimenting with monetizing posts on X (formerly Twitter), and I saw the now famous announcement that Elon Musk had purchased 100,000 GPUs from Nvidia for his AI initiative. That took me straight back to my time as an electrical engineering student at Wichita State University. I guess I internalized the idea that a single custom processor will outperform any attempt to connect many networked processors to solve the same problem. At this point I realized there is a severe gap between what the world wants and needs, and what the silicon industry has to offer. Thus began my deep dive into understanding AI at its core. Once I understood that the core AI problem is a nasty matrix multiplication that is running trillions and trillions of times over a massive data set, I began to wonder if there was an opportunity to improve what I saw.
What makes your product stand out?
The pure integer math foundation of my solution. I speculated that the core of the problem with monstrous matrix multiplications in silicon is the reliance on floating-point encodings. They are slow, energy-intensive, numerically wasteful, and don't scale in silicon that well. I decided if I were going to have a stab at improving things, I would reject the use of floating-point entirely -- even for training AI which heavily relies on floating-point. If I couldn't dream up a solution to avoid floating-point entirely, then I probably had little to offer. So I began twiddling with the underlying math of the entire AI foundation to see if I could invent a pure-integer math solution, and after a few experiments a solution that looked viable evolved.
What have you learned so far on your journey?
The silicon industry is very walled off. It is difficult to reach silicon experts and you must have very deep pockets to even get access to the basic tools it takes to analyze advanced chipsets, let alone design them. But I'm no quitter.
How are the ACCEL-KS grant funds helping you reach the next stage of development?
First and foremost, it was super uplifting. An unspoken secret is that startup founders face doubts early and often, and having that type of recognition is just an adrenaline shot to the arm. Secondly, it has opened up lines of communication that didn't exist prior. Third, I have been self-funding and those costs can add up. The funds are an accelerant that are helping me prove or disprove my concept more quickly.
What kind of support do you need from the startup community?
Connections! I need access to silicon experts, and immediately after that investors.
What is something interesting we should know about you or your project?
If this project succeeds, Tenfold brings these closer to reality: Cancer cures: Every patient is unique, and their cancer is a moving target. From AI's perspective, each individual is a very intense compute problem we just don't have the silicon to solve quite yet. Tenfold advances AI compute by making it dramatically faster and more energy-efficient than what is possible today, bringing hope one step closer. Planes, cars, spaceships can learn: HAL 9000 in 2001: Space Odyssey becomes real. Currently devices with AI can use pre-trained data and make decisions, but having them learn on the fly is a no-go. AI on the desktop: Not toy AI, but advanced models that can sit right in your own computer without even needing Internet connectivity.
