Diogo Almeida is the Co-Founder and CEO of TypeSafe AI, a frontier AI lab building machine-native, composable AI designed to run reliably inside software systems without constant human supervision.

With a decade of deep AI research experience, Diogo previously worked at OpenAI and Google Brain, where he co-invented RLHF and InstructGPT, the foundational methods that led to ChatGPT and GPT-4.

Diogo argues that RLHF, while transformative for human-facing AI, constrains what AI can accomplish independently, and that most intelligence should eventually run quietly inside software rather than interact with people.

At TypeSafe AI, he leads development of a new model architecture called Reinforcement Learning for Calibrated Decisions (RLCD), powering their debut model Jev, which delivers frontier-level intelligence at under 100 milliseconds of latency.

TypeSafe AI emerged from stealth in September 2026 with $40 million in seed funding led by DCVC, offering a model up to 100 times faster and less expensive than other frontier models for developers building autonomous software systems.