TypeSafe AI Inc., a startup founded by a former OpenAI Group PBC researcher who helped develop ChatGPT, emerged yesterday with $40 million in seed funding and a model designed to put artificial intelligence directly inside software applications.
The San Francisco-based company says its first model, called Jev, differs from conventional large language models by producing structured decisions rather than conversational text. The goal is to give developers a faster and more predictable AI component for tasks that require judgment but can’t be handled easily with conventional rules.
Deep technology venture capital firm DCVC Management Co. LLC led the round. Forbes reported that the firm valued TypeSafe at $200 million, citing a person familiar with the transaction. The company was founded in 2024 by Chief Executive Diogo Almeida, who previously worked on reinforcement learning from human feedback, InstructGPT, ChatGPT and GPT-4 at OpenAI. Co-founders Erik Gafni and Sasha Sheng also have extensive AI experience.
TypeSafe’s thesis is that the qualities that make AI models effective at talking to people can work against them when they are incorporated into production software. Large language models can generate plausible but incorrect information, vary their methods between requests and present uncertain answers confidently. Applications that depend on consistent outputs therefore often require human review.
“We’ve been optimizing for humans, and we’re superhuman at pleasing humans,” Almeida told Forbes, referring to popular generative AI models.
Machine-ready output
Jev instead accepts structured questions and returns typed answers that other software can process. Those answers can take the form of yes/no probabilities, selections from a defined list or scores on a specified scale. Each decision includes probabilities and a confidence measure, enabling developers to set thresholds for when an application should proceed automatically, seek more information or refer a decision to a person.
In insurance underwriting, for example, Jev could review available evidence about a property and estimate the likelihood that it will catch fire. A developer could let an automated workflow continue when confidence is high, while routing an ambiguous case to an underwriter.
TypeSafe calls Jev a “System One Model” trained with a method it calls Reinforcement Learning for Calibrated Decisions. The approach is intended to turn AI into a composable software primitive that can be combined with deterministic code. Hundreds of decisions can be generated in parallel from one prompt and assembled into larger workflows, according to the company.
The company says Jev uses parallel processing to deliver results in less than 100 milliseconds, making it up to 100 times faster and less expensive than other frontier models. Its website lists a cost of 39 cents per 1,000 workflows, compared with $3.31 for OpenAI’s Gpt-5.6 Luna and R19.49 for Anthropic PBC’s Claude Haiku 4.5. TypeSafe reported that its own tests found Jev to be nearly 194 times faster and about 445 times cheaper than the language models used for comparison. Those figures have not been independently verified and will vary by workload, network location and comparison method.
Jev’s constrained output doesn’t guarantee correct decisions, but the company said its use of probability ratings minimizes the risk of hallucinations, in which models present the wrong answer with a high level of confidence. Developers need to test how well Jev’s confidence scores correspond to actual accuracy on their own data.
TypeSafe AI said its largest opportunity is in high-volume business processes such as classifying service requests, evaluating invoices, triaging security alerts and reviewing the results of AI agents. In such cases, a specialized decision model could serve as a control layer while a language model handles tasks such as drafting text.
DCVC General Partner James Hardiman said TypeSafe is addressing “one of the biggest remaining challenges in AI” by making models reliable enough to embed in products at scale.
Jev is currently available through an early-access waitlist.
