Arturo Avelar
Lately I've been thinking about structures and conventions, and how they define what's possible.
Thesis
I look for places where challenging assumptions or rethinking foundations make systems work better. Complexity isn't a problem by itself, some of the most interesting systems are complex by nature. But in companies, products, or industries, it often shows up when foundations are fragile.
In those cases, inefficiencies tend to be signals that something underneath hasn't been properly structured.
Exploration
Building
Working on Valar, trying to make business insurance more accessible, understandable, and tailored for SMEs across LATAM. A lot of it comes down to redesigning distribution, improving transparency in contracts, and making risk easier to understand for carriers.
Thinking
About the rails insurance will run on for the next 100+ years. The current ones feel a bit rusty.
Learning
Spending time on AI, physics, and robotics. Not directly related to insurance, but useful for thinking about systems, constraints, and how complexity actually behaves. Also just enjoyable.
Work
Valar
I am working on the underlying infrastructure of commercial insurance. Most of the friction does not come from pricing or underwriting itself, but from how the system is structured: fragmented distribution, opaque contracts, and repeated reinterpretation of the same information.
The goal is to define infrastructure where information can move across brokers and carriers without being rebuilt each time.
How policies are represented. Policies are usually static documents, but the underlying risk changes over time. I have been exploring ways to treat them more like data so they can be compared and reasoned about.
How distribution works. Much of the system depends on intermediaries and fragmented communication. I am interested in what new distribution layers could look like, including embedded models and API-based access.
How risk is modeled. From underwriting to parametrics, I keep coming back to how models depend more on the structure of the input than on the model itself.
PDF to structured data
Working on an open source pipeline to convert PDFs into structured, usable data using OCR and LLMs.
Most documents are designed to be read once, not reused. I am interested in turning them into something that can be queried, transformed, and integrated into other systems.
Teaching
Teaching forces a different kind of clarity. Concepts that feel intuitive when building have to be broken down into something that can be explained and rebuilt from first principles.
It has changed how I think about structure, especially in how ideas are introduced and connected.
Thinking
Most inefficiencies are not problems to solve, but signals pointing to something deeper.
A good model can't fix badly structured input.
The hardest part is not building the system, but deciding what the system actually is.
A lot of “complexity” is just accumulated decisions that were never revisited.
Policies are read once and then forgotten, even though they define everything that follows.
Good abstractions remove work, not add features.
Most systems break at the boundaries, not at the center.
Closing
Always building, trying to understand how systems actually work.