Diogo Almeida is an AI researcher, entrepreneur and the co-founder and chief executive of TypeSafe AI, the company behind the Jev decision model.
Before starting TypeSafe, Almeida worked on research that helped shape the modern chatbot era. His company credits him as a co-inventor of reinforcement learning from human feedback and InstructGPT, the research line that made language models better at following instructions and directly influenced ChatGPT.
What is Diogo Almeida known for?
Almeida is best known for two connected parts of his career.
First, he worked on training methods that helped make large language models more useful to humans.
Second, after that work, he founded a company arguing that many software problems should not use a chatbot at all.
That second idea became TypeSafe AI and its first model, Jev.
Did Diogo Almeida work at OpenAI?
Yes.
TypeSafe’s official biography says Almeida worked on the research behind RLHF and InstructGPT, two major steps in the development of instruction-following language models.
RLHF became one of the most widely discussed techniques in modern AI because it uses human feedback to shape model behaviour after the initial training process.
What did he do before OpenAI?
TypeSafe says Almeida previously worked at Google Brain.
Google Brain was one of the major research groups that helped push deep learning forward before Google consolidated more of its AI work under the Google DeepMind organisation.
Why did he start TypeSafe AI?
Almeida has said that one question kept bothering him after the success of chat models: if the models are already extremely capable at conversation, why has software automation not improved at the same speed?
His conclusion was that text generation is not the right interface for every kind of intelligence.
Software often needs a decision, category, probability or action. It does not need a beautifully written paragraph before every branch in the program.
What is TypeSafe AI?
TypeSafe AI is the company Almeida co-founded with Erik Gafni and Sasha Sheng.
The company spent roughly two years in stealth before releasing its first public model in September 2026.
Its focus is what it calls System One models: models built for structured decision-making rather than open-ended text generation.
What is Jev?
Jev is TypeSafe’s first public model.
Instead of generating free-form language, it takes unstructured input and returns a typed decision with a probability.
That makes it useful for tasks such as routing, classification, automation and software control.
See our Jev explainer for how the model differs from a normal LLM.
Why move away from the technology he helped create?
Almeida is not arguing that chat models are useless.
His thesis is that humans should not be the only consumers of machine intelligence. A large amount of AI work can happen invisibly inside software where the output is meant for another program rather than a person.
That is why TypeSafe is trying to optimise directly for decisions that code can consume.
How much has TypeSafe raised?
The company has announced US$40 million in seed funding led by DCVC.
Recent reporting says investor interest has increased sharply since Jev launched, with discussions around a much larger financing and a possible valuation above US$10 billion.
Those larger figures refer to reported talks, not a completed financing announcement.
Why is Almeida getting attention now?
Jev has become one of the most talked-about alternative AI architectures in Silicon Valley since its September launch.
The unusual part is the founder’s background: Almeida helped create techniques that made chatbots successful and is now betting that a huge part of the next AI market will sit outside chat.
Does TypeSafe say Jev cannot hallucinate?
The company argues that Jev avoids ordinary text hallucination because it does not generate open-ended text.
A decision model can still make an incorrect prediction, so “no hallucination” should not be interpreted as “never wrong.”
What else is known about him?
TypeSafe’s official team page includes a more personal detail: Almeida has competed at a very high level in games, reaching the top ranks in League of Legends and Hearthstone.
We have left out age, family details and wealth estimates because reliable primary sourcing for those details is limited and they do not help explain his work.
Why his career matters
Almeida’s career connects two generations of AI product thinking.
The first generation was about making language models better at talking to people. The second question he is now exploring is how to put intelligence directly into software without requiring a human conversation at every step.
For more founder and technology profiles, see Holyprofweb’s People section.