Jev founders
Jev founders are the TypeSafe AI founders who shipped Jev, the first public System One model: Diogo Almeida (CEO), Erik Gafni (CTO), and Sasha Sheng (COO). Bios below are compressed from public pages — official team, the 15 September 2026 launch post, TechCrunch (18 September 2026), and the BusinessWire seed note. If this page and the team page disagree, the team page wins.
This page is for builders who want the research/ops story behind RLCD and System One — not a fan wiki and not TypeSafe HR. Company envelope: TypeSafe AI. This guide’s editors: about jev.pro.
Why founders matter for the mental model
Jev is easy to misfile as “a small LLM” or “just a classifier.” The public founder story is the corrective:
- Almeida’s published complaint is that chat-optimized models (RLHF / human language) are a poor interface for software automation.
- TypeSafe’s named method is RLCD — calibrated decisions with probabilities the surrounding code can gate on — not “write a nicer paragraph.”
- Gafni’s public bio is production AI systems; Sheng’s is research engineering + shipping culture. That pairing is why the docs obsess over schema, confidence, and jaggedness instead of a chat window.
You still have to measure your questions. Pedigree is not a benchmark. Jev review.
Diogo Almeida — CEO
Official team blurb: co-invented RLHF and InstructGPT, “the methods that lead to ChatGPT and GPT4”; previously Google Brain.
Launch post byline (15 September 2026): Almeida writes that at OpenAI he helped build instruction-following / chat methods that “ended up as the research behind ChatGPT,” then concluded something large was missing for automation. Jev is framed as a frontier-intelligence function call.
TechCrunch (18 September 2026) quotes him:
“We have lightning in a bottle, and yet it is not useful… The problem is we are optimizing for human language.”
And on synthetic data:
“We made an early bet that we will be making all of our data, and that has been one of the best bets I’ve ever made in my life — better than our launch, in my opinion, better than RLHF.”
The same article says he describes training as reinforcement learning from calibrated decisions (TypeSafe’s docs spell RLCD). Architecture details stay unpublished; we do not fill that gap.
Team-page fun fact (theirs): top-30 League of Legends NA / top-2 Hearthstone. Color, not a buying criterion.
Erik Gafni — CTO
Official team blurb: repeat founder (Ravel, multi-modal AI for DNA sequencing); early employee at Invitae and Freenome; inventor with publications and patents; “specializes in building production AI systems.”
That is the builder-facing half of System One: latency envelopes, pinning, traces, and “put the decision in the request handler.” It is also why jev.pro’s operate cluster exists — production craft is the product, not a demo notebook.
Team-page fun fact (theirs): 44th at US Chess Nationals in elementary school.
Sasha Sheng — COO
Official team blurb: ex-research engineer at Meta / FAIR (News Feed, AI Experiences, AI Research); hackathon organizer; publications at NeurIPS and ECCV.
Team page: “Won 7 hackathons in 2023” plus the caffeine/Barry’s line. The useful signal for practitioners is ops + research-engineering, not the fun fact: waitlist, console, and the gap between a launch graph and a on-call runbook.
One founders page, not three thin bios
Almeida has the longest public interview trail. Gafni and Sheng’s public pages are short. Splitting /diogo-almeida-typesafe style URLs would duplicate this page without new intent. Prefer typesafe.ai/team for the live roster and LinkedIn buttons.
BusinessWire (15 September 2026) names all three as founders alongside the ~$40M DCVC-led seed. Details: TypeSafe AI.
FAQ
Are these people the jev.pro editors?
No. TypeSafe’s team is TypeSafe’s. This site is independent — about jev.pro.
Is Jev named after a founder?
No. TypeSafe’s launch FAQ: Jev after William Stanley Jevons (Jevons paradox). System One after Kahneman’s System 1.
Can I email founders through this site?
No. Use TypeSafe’s published contacts ([email protected] / [email protected] on their properties).
What this page does not claim
- No private bios, ages, or equity splits.
- No “the team is why Jev scores X.”
- We do not impersonate TypeSafe or recruit for them.