Meet Diogo Almeida, former OpenAI researcher behind Jev, an AI model designed to deliver structured answers, scores and yes-or-no decisions

Diogo Almeida, a former researcher at OpenAI, has launched a new artificial intelligence (AI) model called Jev. It is is designed to deliver structured answers, scores and yes-or-no decisions instead of generating text like conventional chatbots. TypeSafe AI, a startup led by Almeida, has developed Jev. The model had attracted attention in Silicon Valley as companies explore alternatives to traditional large language models (LLMs).

Unlike chatbots that generate text in response to prompts, Jev classifies information into predefined outcomes. these can include a yes-or-no answer, a numerical score or a choice from a fixed list of options.

The startup is betting that this approach can make AI faster, cheaper and more predictable for software automation.

What makes Jev different?

Jev uses a machine-learning approach that TypeSafe calls "Reinforcement Learning for Calibrated Decisions." Rather than generating a sequence of words, the model is designed to make decisions within a defined set of possible answers.

This could make it useful in software systems that need to process large numbers of requests and make repeated decision without requiring a person to review every result.

Almeida has argued that traditional LLMs can be difficult to use for reliable automation because their outputs may be unpredictable. Jev is designed to provide structured results that software can use directly.

Why did Almeida develop Jev?

Almeida left OpenAI in 2024 and spent about two years developing the idea behind TypeSafe AI. According to WSJ report, Almeida believes that although existing AI models ca perform complex tasks with human involvement, they have limitations when used to build fully automated software systems.

Traditional chatbots generate responses in natural language, which software may then need to interpret. A model that returns predefined decisions can reduce the need for that additional processing.

TypeSafe AI emerged from stealth with Jev's launch and announced that it had raised $40 million in funding from venture capital firm DCVC, according to WSJ report.

Almeida told The Wall Street Journal that Jev was already being used by about 25% of Fortune 500 companies. He also said the model was processing around one trillion tokens per day about a week before the interview.

The company is positioning Jev as a tool for developers who want to integrate AI into automated software. Its focus is on making repeated decisions quickly rather than producing lengthy written responses.

Are other companies developing similar tools?

Jev's launch has coincided with the introduction of similar products from established AI and technology companies.

On September 29, OpenAI introduced its Decisions API, which allows developers to use its Luna model to answer specific questions from a predefined set of possible answers.

The following day, data analytics company Databricks unveiled a function called ai_decide. The company said demand was growing for fast, low-cost tools that turn raw text intro structured decisions.