The Harumi Thesis: From Luck to Determinism

I had just left my previous startup after some misalignments with my former cofounders.

Around that time, Antonio Brennand, an angel investor from one of Brazil's wealthiest families, invited me to attend some mathematics classes at Prandiano Museu da Matemática in São Paulo. The classes were taught by Aguinaldo Ricieri, the founder and owner of Prandiano and a professor from Instituto Tecnológico de Aeronáutica (ITA).

One of the topics covered in those classes was Applied Mathematics, specifically the field of Operations Research. Ricieri works as an Operations Research consultant for large companies and described the manual process he used to solve operational problems for his clients.

—

Around the same time, I was reading The Almanack of Naval Ravikant.

One idea from the book stayed with me: luck isn't entirely random. Naval describes four types of luck and argues that, while you can't control whether opportunities appear, you can systematically increase the odds of recognizing and capitalizing on them through preparation, relentless action, and by becoming uniquely good at something. In other words, luck isn't just something that happens to you, it can be engineered.

I was looking for a way to maximize my odds of creating value and building a successful company. If my startup helped already-successful businesses become even more successful, maybe I was moving the "luck factor" from being dependent on me to being dependent on my customers, thus multiplying my chances of becoming rich.
In other words, and also using a metaphor shared by Naval, others could have found a treasure deep in the ocean and I would be the one with the diving expertise to help my clients reach the treasure and charge a percentage of the value I helped unlock. Instead of searching for my own treasure and depending on luck, I could help people who had already found one but couldn't reach it. My job would simply be to become the best diver.

—

At first, studying Operations Research and reading The Almanack of Naval Ravikant felt like two completely unrelated experiences. Eventually, I realized they were pointing to the same conclusion.

During the math classes, I realized that Operations Research could be my next thesis: Operations Research helps companies improve the efficiency of their operations through mathematical modeling and simulation. These companies had already found their treasure. My job would be to help them reach it: I wanted to eliminate luck from entrepreneurship as much as possible. Rather than betting on discovering the next billion-dollar market, I wanted to become exceptionally good at creating value wherever it already existed. It felt almost deterministic.

As I thought more about it, other pieces started falling into place:

• Strong founder-market fit: I genuinely enjoy mathematics.
  
  

• Little competition: Most people working in Applied Mathematics and Operations Research pursue careers in academia or consulting rather than building startups.
  
  

• A broken delivery model: Companies need this type of solution, but the traditional delivery process is almost entirely manual. Operations Research engineers build mathematical models, software engineers translate them into production systems, front-end engineers create interfaces for business users, and infrastructure teams build the platform required to execute millions of optimization scenarios. The amount of specialized work involved makes projects slow, expensive, and inherently difficult to scale.

—

The thesis felt right, but I still hadn't found the person capable of turning it into a scalable product. That changed when I met Marcel Nicolay.

Marcel had been Product Director at Gympass and previously co-founded a YC-backed startup as its CTO. Marcel immediately saw the same opportunity and, more importantly, knew how to build the technology required to make it scalable.

Together, we interviewed Operations Research experts to understand how they worked and where the biggest bottlenecks were. Those conversations shaped the first version of Harumi so that it could use AI to accelerate all the steps of delivering an OR project to our clients.

—

Looking back, what looked like a series of unrelated events (math classes, Naval's book, meeting Marcel, interviewing Operations Research experts) turned out to be one coherent path. I just couldn't see it at the time.

Whether this thesis proves right is something time will tell. But it became the foundation upon which we decided to build Harumi.