01 / BRIEF
A microlending fintech for the tens of millions of Brazilians without access to traditional credit. Co-founded as CTO, from an empty repository: no product, no schema, no team, no first line of code.
Being invisible to a credit bureau is not the same as being a bad borrower. The entire company is built on that distinction, and everything downstream of it, the model, the flow, the infrastructure, had to be designed rather than adopted, because the off-the-shelf versions all assume a customer who already has a file.
02 / HOW IT WORKS
Greenfield, product first. Before any architecture there was a question: what does a loan look like for someone who has never had one. The answer shaped the product. Small first amounts, a flow that explains itself while it runs, questions asked in an order a person can actually answer, and a clear no when the answer is no. A rejection that feels arbitrary costs you the customer forever, including the version of them that would have qualified next year.
Greenfield, software second. A TypeScript API, a React web app, and Postgres underneath, running on managed cloud infrastructure so a team of five never has to babysit a server. Payments ride the country's instant rails rather than the banking system's business days, which is the difference between credit that arrives when it is needed and credit that arrives after the problem.
Scoring a thin file. The person who needs credit most is the person the system knows least about. Scoring someone with ten years of bank history is a solved problem; scoring someone the banks never looked at requires different signals, a model built for that shape of data, and the humility to start with small amounts. Every repayment goes back into the model, so the book teaches the lender about exactly the population nobody else has data on.
The phone is the platform. Every screen assumes a low-end Android on a network that drops mid-request. That is not an edge case to handle later, it is the median user, and designing for it first made every other decision simpler.
03 / DECISIONS
Own the risk model. Buying a score off the shelf means inheriting somebody else's blind spot, and their blind spot is precisely our customer.
Instant rails, not bank days. If the money lands three business days later, the emergency it was for has already been solved some worse way.
Start small on purpose. A first loan is a question, not a bet. Small amounts let the model be wrong cheaply, and let a good borrower prove themselves fast.
Boring infrastructure, on purpose. Managed services over clever ones. In a company this size, every hour spent operating something is an hour not spent on the product that has to convince a stranger to trust it with money.
Say no clearly. A vague rejection is a product decision, and a bad one.
04 / STATUS
Shipped 2024. Ongoing.