The TypeSafe Founders
Three co-founders, one of whom helped invent the training method his new company was built to move past. That tension is the most interesting thing here.
Diogo Almeida
Co-founder and CEODiogo Almeida was formerly at OpenAI, where he worked on reinforcement learning from human feedback and early ChatGPT research. He is credited as a co-inventor of RLHF — the method his new company's RLCD is explicitly positioned against.
Erik Gafni
Co-founderNamed among the TypeSafe founders in the launch announcement. TypeSafe published little biographical detail beyond Diogo Almeida, and no complete roster appears in the launch blog.
Sasha Sheng
Co-founderNamed among the TypeSafe founders in the launch announcement. As with Gafni, the company's own launch post centres the method rather than the team, so outside reporting is the only source for the full founder list.
Diogo Almeida, an RLHF co-inventor against RLHF
The most interesting fact about the TypeSafe founders is a biographical one. Diogo Almeida is credited as a co-inventor of reinforcement learning from human feedback — the technique that made language models useful enough to become a consumer product. The training method Diogo Almeida now sells, RLCD, is defined in opposition to it.
The Diogo Almeida distinction is not that RLHF was a mistake. It is that RLHF optimises for a specific thing: responses a human rater prefers. That is exactly right when the consumer is a human reader, and exactly wrong when the consumer is a program that needs a number it can threshold on. Preference is about how an answer reads. Calibration is about whether the stated confidence survives contact with the outcome.
Coming from Diogo Almeida, who built the first thing, that critique carries weight it would not carry from a competitor. It is a narrower claim than it sounds: not that chat alignment was wrong, but that it was solved for the wrong consumer, and that nobody went back to solve it for the other one.
TypeSafe founders timeline
The August repository date is a small but telling detail about how the TypeSafe founders planned the launch. Publishing a client adapter five weeks ahead is not an accident — it means the integration surface was frozen well ahead of the launch, and that developer adoption was being planned for rather than hoped for.
What the TypeSafe founders did not disclose
Beyond the three TypeSafe founders, very little. There is no published headcount, no research team roster, no list of advisors, and no biographical material on the two co-founders who are not Diogo Almeida. For a company whose central technical claim is a novel training method, the absence of a named research team is a gap — it is the kind of thing that normally accompanies a methods post.
That is not evidence of anything untoward. Two years in stealth followed by a deliberately staged launch tends to produce exactly this shape: a strong Diogo Almeida story, a named method, and the details held back for later material. But it does mean that the usual way of evaluating a frontier lab — reading the author list and following the citations — is not available here yet.
The thing to watch is whether the TypeSafe founders publish a technical paper. A described method with no published evaluation is a product claim; a described method with a paper and reproducible results is a contribution. Which of those this turns out to be will say more about the TypeSafe founders than the funding round did.
Why the Diogo Almeida pedigree matters commercially
Frontier labs are hard to evaluate from outside. There is no revenue to inspect, no product to try without an invitation, and no published architecture to judge. In that vacuum, the founder track record does a lot of the work — and a Diogo Almeida byline on the RLHF lineage is about as strong a signal as the field offers.
That is presumably part of what a $40M seed is buying. Investors backing the TypeSafe founders at this stage are backing a claim about a training method that has no paper, on a model with no third-party evaluation, in a category that did not exist six months ago. The Diogo Almeida history makes that legible as a bet rather than a leap.
It cuts the other way too. A pedigree that strong sets expectations the TypeSafe founders now have to meet, and the standard for a lab of this provenance is a reproducible method rather than a rate card. The absence of a research team roster alongside the Diogo Almeida story is the gap most worth watching over the next year.