Japan's Offers platform added Jev experience to profiles and employer searches across a network it says includes 40,000 professionals.
By [RuntimeWire Staff](https://runtimewire.com/author/runtimewire-staff)
· Published
Primary source: [Aligned News - AI Intelligence](https://x.com/polarbrowser/status/2101730142355857868)
Why it matters #
Offers is treating familiarity with Jev as a searchable professional skill just five days after the model's early-access release. The listing lets employers find candidates who report Jev experience, while stopping short of automated candidate judgment.
On September 20th, Offers, a Japanese recruiting platform, announced that it had added experience with TypeSafe AI's five-day-old Jev model as a profile skill and a condition employers can use when searching for candidates or sending offers.
Jev is the first public model from TypeSafe AI, founded by Diogo Almeida, Erik Gafni and Sasha Sheng. The company released Jev in early access on September 15th.
Offers, operated by overflow, says roughly 40,000 professionals have registered on the platform and that it has supported hiring at more than 1,000 employers since its 2019 launch.
A brief X post billed the development as "Jev for Hiring." The primary announcement is narrower. Offers added Jev to its skill taxonomy; Offers has not said that Jev is screening applicants, ranking candidates or making employment decisions.
The supplied public materials do not disclose TypeSafe AI revenue, customer or production-adoption figures.
Almeida's bet against chat as the default interface
The hiring update follows TypeSafe AI's September 15th launch of Jev, the first model in a category TypeSafe AI calls "System One Models." TypeSafe AI spent about two years in stealth before releasing Jev in early access.
Almeida's founding thesis came from his previous work on conversational models. TypeSafe AI identifies Almeida as a former Google Brain and OpenAI researcher who helped develop reinforcement learning from human feedback and InstructGPT. In TypeSafe AI's launch post, Almeida wrote that models had become highly capable at chat while widespread automation remained elusive. He concluded that software needed a different interface to intelligence.
Jev accepts unstructured application state, such as text, and returns predefined, typed decisions with probabilities and confidence scores. A developer might use it to classify an incoming request, route a support ticket, score a record or decide which branch of a workflow should run. Jev does not generate free-form prose for those tasks.
That design reflects Almeida's bet that much of the economic demand for AI will come from frequent, bounded decisions inside ordinary applications. Software can consume Jev's output directly without parsing a paragraph and hoping the model followed the requested format.
TypeSafe AI launched with a $40 million seed round led by DCVC. A seed of that size gives Almeida and his co-founders room to pursue a new model architecture, though it also sets a high bar for proving that cheaper machine-readable decisions translate into durable usage.
A founding team built around production systems
Gafni brings TypeSafe AI's clearest history of building machine-learning systems outside major research labs. On his personal site, he says he worked on bioinformatics research at Harvard Medical School, joined genetic-testing businesses Invitae and Freenome early, and co-founded cancer-screening developer Ravel Biotechnologies, which raised $9.5 million. He later led machine-learning research at Eventum AI.
Sheng previously worked as a research engineer at Meta and FAIR, according to TypeSafe AI. Her work covered News Feed, AI experiences and research, and TypeSafe AI says she has published at NeurIPS and ECCV. Together, the founders have assembled a San Francisco team with experience from OpenAI, Google Brain, Meta, Stripe, Airbnb, Plaid and Docker. A TypeSafe AI job posting describes the company as fully in person at its San Francisco office.
The backgrounds fit the product. Jev is being sold as infrastructure for production software, where latency, predictable schemas and confidence calibration matter alongside raw model capability. Gafni's history in diagnostics also gives the founding group experience with machine-learning systems whose outputs must be integrated into larger operational processes.
A hiring credential is still several steps from hiring software
Offers added Jev to its skill taxonomy, allowing candidates to list experience with the model and employers to search for it. Offers has already added emerging AI roles and skills to its profiles, including forward-deployed engineering and deployment strategy.
Offers has added Jev to its skill taxonomy, allowing candidates to list experience with the model and employers to search for it. Neither side has disclosed how many candidates have added the skill or how many employers have used it.
Five days is also too short a window for Jev experience to represent a deep professional specialization.
In its launch post, TypeSafe AI says Jev's end-to-end response time is 70 to 500 milliseconds, input pricing is $0.042 per million tokens and output tokens are free. TypeSafe AI says Jev can run 40 to 200 times faster than large language models on tasks shaped for its structured approach. Those comparisons come from TypeSafe AI's own evaluations. TypeSafe AI acknowledges that its published workflows were prepared internally and may contain bias, and independent evidence for Jev's accuracy, calibration and performance in hiring-related workflows remains limited.
Offers' announcement avoids making those broader claims. It recognizes Jev knowledge as a credential without placing the model in charge of candidate assessment. That is a sensible first step for technology touching employment: establish whether developers use the model before asking it to judge the people who do.