TypeSafe launches Jev model for programmatic logic

TypeSafe, an artificial intelligence startup founded by a ChatGPT co-inventor, has emerged from stealth mode to launch its Jev model aimed at automating programmatic decisions using a parallel sampling architecture. The release allows software systems requiring automated deterministic logic to bypass conversational language models entirely by executing structured probabilistic decisions directly inside production codebases.

Developed during a two-year stealth period by OpenAI veteran and ChatGPT co-inventor Diogo Almeida, the model discards text and string generation completely. Instead of autoregressively generating sequential text tokens, the system takes an unstructured state as input and outputs type-safe structured values within a single parallel query, targeting deterministic code integration and eliminating output hallucinations by design.

"Think of Jev as a frontier-intelligence function call: unstructured state in, typed probabilistic decisions out."

Diogo Almeida, Founder at TypeSafe

Engineers constructed the platform around an alternative training methodology termed Reinforcement Learning for Calibrated Decisions (RLCD). Unlike conventional frontier models trained via RLHF or RLVR to satisfy human conversational preferences, TypeSafe trains Jev to produce calibrated probabilities on execution logic, ensuring that returned confidence scores correlate directly with output accuracy.

Departing from traditional autoregressive token generation, Jev relies on a hardware-aware parallel sampler that evaluates and delivers all structured values simultaneously. This structural constraint restricts outputs to predefined schemas, removing the parsing pipelines and guardrail wrappers usually required when deploying large language models across business automation workflows, while handling high-cardinality selections up to 255 discrete options.