For years, large artificial intelligence models were trained primarily on texts, images, and other content available on the internet. Now, companies like OpenAI and Anthropic are seeking a type of data that is harder to find: the knowledge accumulated by professionals throughout their careers.

This is the market in which the Brazilian company Vetto AI operates. The startup recruits lawyers, engineers, programmers, financial analysts, and other specialists to transform their experience into data used to train the AI.

“Human knowledge isn’t on the internet or in a book. It’s in someone’s daily work,” says Lucas Smaira, co-founder and head of research at Vetto, in an interview with Revolução IA, a NeoFeed program supported by Casas Bahia.

The company acts as a bridge between specialists and the laboratories responsible for the most advanced models on the market. Professionals work remotely, at their preferred times, and are paid per task. Depending on the complexity and specialization required, payment can reach US$300 per hour or even exceed that amount.

The argument is that, although models can already access and process large volumes of information , they still depend on humans to learn decisions that are not objectively described. In a company valuation, for example, AI can calculate metrics, but an expert still needs to indicate which companies make sense as a benchmark.

Vetto combines this knowledge with synthetic data, produced by the artificial intelligence itself. This allows humans to focus on insights that the model cannot yet generate, while repetitive steps are automated.

Founded in November of last year, Vetto has already become profitable, Smaira said, driven by contracts with major artificial intelligence laboratories. This business model, however, presents a challenge: since the market only has five or six potential global clients, losing a significant contract could mean a 20% to 30% reduction in the startup's business.

Because of this, this year Vetto began operating in another area, the corporate market , expanding its client base. In this case, the startup helps organizations measure the efficiency of their adopted models and reduce expenses related to artificial intelligence, avoiding the unnecessary consumption of tokens.

According to Smaira, many companies use sophisticated and expensive systems even for simple tasks, which is a "waste".

“I would guess that more than 80% of the tokens are useless, they don’t generate value, and we’re paying for them,” he said. And he compared it to: “You wouldn’t want to use a Ferrari to drive on the cobblestones of a street in São Paulo.”