Who Is Jared Palmer, the Creator of Kev and Turborepo?

Jared Palmer is the vice president (VP) of engineering at Cognition, the applied artificial intelligence (AI) laboratory behind the Devin coding assistant, and the creator of the Kev open-weight decision model family.

At Layer3Labs, we build AI automations and agentic workflows for small and mid-sized businesses, so we evaluate developer tooling and decision architectures through their production reliability.

Before Cognition, Palmer created the open-source projects Formik and TSDX and founded the build system Turborepo. He was later VP of AI at Vercel, where he created v0, and then held product and engineering roles at Microsoft, GitHub and Xbox.

Who Jared Palmer Is

Jared Palmer is a software engineer who serves as VP of Engineering at Cognition, having previously founded Turborepo and created Formik, TSDX, and the v0 application. On 1 October 2026, Palmer released Kev 1.0, an open-source family of decision models built to output calibrated probabilities over structured choices in one forward pass. He joined Cognition in July 2026 after serving as VP of Engineering at Xbox and holding executive product positions across Microsoft and GitHub.

KitGuru and Outlook India Respawn covered his July 2026 move from Xbox to Cognition, about three months after he took the Xbox role. More profiles are in our list of AI people to watch.

• Current role: VP of Engineering at Cognition (leads engineering), per his bio.
• Organization: Cognition, the applied AI laboratory behind the Devin coding assistant.
• Known for: Creating Formik, TSDX, Turborepo, the Vercel AI SDK, and v0.app.
• Previous roles: VP of Engineering at Xbox; VP of AI at Vercel.
• In the news for: Releasing Kev 1.0, an open-source decision model family, on 1 October 2026.

Formik, Turborepo and Vercel

Jared Palmer's bio says he earlier ran a software development and design studio in New York City. The bio gives no dates for the studio.

The Vercel blog post announcing the Turborepo acquisition described Palmer as the creator of the open-source projects Formik and TSDX.

Jared Palmer founded Turborepo, a build system for JavaScript and TypeScript codebases. On 9 December 2021, Vercel announced its acquisition of Turborepo, and Palmer joined Vercel.

At Vercel, Palmer was VP of AI and created v0.app and the AI SDK, according to his bio. The same bio describes v0.app as "one of the world's most popular AI coding agents."

Product Leadership Across Microsoft, GitHub, and Xbox

In October 2025, Jared Palmer left Vercel to join Microsoft's CoreAI organization. A November 2025 episode of the Syntax podcast documented his transition, titled 'Why v0 creator left Vercel to fix GitHub.' Reporting from KitGuru noted that he held the titles CoreAI VP of Product and Senior VP at GitHub, while Palmer's about page records his role as SVP of Product at GitHub.

In May 2026, Palmer transferred within Microsoft to become VP of Engineering at Xbox, reporting directly to Xbox chief executive officer (CEO) Asha Sharma, according to Outlook India Respawn.

Palmer left Xbox in July 2026 after approximately three months in the role, as reported by KitGuru. In his departure announcement on social media, Palmer wrote: 'I'm grateful to [Asha Sharma] and everyone at Xbox for an amazing year.' His exit concluded a nine-month tenure across Microsoft's CoreAI, GitHub, and Xbox divisions.

Cognition Engineering Leadership and Devin

On 21 July 2026, Jared Palmer announced on X that he had joined Cognition to lead engineering, as reported by Outlook India Respawn. His personal site lists his title as VP of Engineering at Cognition, leading engineering teams for the applied AI laboratory.

For the engineer who created Claude Code at Anthropic, see our profile of Boris Cherny.

The Kev blog post and GitHub repository do not mention Cognition or any other employer, and Palmer publishes Kev under his personal GitHub and Hugging Face accounts. Neither source describes Kev as a Cognition product or as part of Devin.

Architecture of the Kev Decision Model Family

On 1 October 2026, Jared Palmer published Introducing Kev and tagged the Kev 1.0 release on GitHub under the Apache-2.0 license. The Kev repository contains four decision models that read a document and typed questions, then return calibrated probabilities over options in one forward pass. Rather than generating conversational text tokens, Kev outputs structured probability distributions over predefined categorical choices.

Kev models are built on Qwen open-weight architectures, spanning four discrete parameter sizes: 0.8 billion, 4 billion, 9 billion, and 27 billion parameters. On X, Palmer described Kev 1.0 as 'an open weight family of decision models you can train and run on your own.' In a follow-up statement on X, Palmer explained that the flagship Kev-27B model is a supervised fine-tuning (SFT) of the Qwen3.8-27B text backbone trained on approximately 146,000 records and 337,000 questions.

Kev-27B has a 65,536-token context window. The 0.8B, 4B and 9B models each have 8,192 tokens, per the Kev repository.

The official Kev repository defines the project as a "Jev-like family of decision models built on top of Qwen3.5/3.8 you can train and run on your own." The repository also says "the TypeSafe Python SDK works against a Kev server unchanged" to support existing client implementations.

• Kev 1.0 includes 0.8B, 4B, 9B, and 27B parameter variants built on Qwen3.5 and Qwen3.8 base architectures.
• The models operate under the permissive open-source Apache-2.0 software license.
• Kev-27B features a 65,536-token context window, while smaller variants support 8,192 tokens.
• Kev models return calibrated probabilities over the options in a single forward pass.

Why Jared Palmer Built Kev

Palmer built Kev as a self-hosted counterpart to TypeSafe's Jev decision model. In his introductory essay, Palmer wrote: 'I wanted that interface with weights I could run and fine-tune myself.' The Kev repository documents that Kev is compatible with TypeSafe's hosted Jev decision model, allowing the TypeSafe Python SDK to run against a self-hosted Kev server without code modifications.

Kev belongs to the category of system one decision models, which return probabilities over a fixed set of options instead of free text. Our Jev vs Kev comparison and Laya vs Kev comparison set Kev against those two models.