Co-founders behind Reface and Prisma join hands to improve on-device model inference with Mirai | TechCrunch

TechCrunch

by Ivan Mehta

February 19, 2026

AI-Generated Deep Dive Summary

Mirai, a London-based startup founded by Dima Shvets (co-founder of Reface) and Alexey Moiseenkov (co-founder of Prisma), has raised $10 million in seed funding led by Uncork Capital. The company aims to improve AI model inference on devices like smartphones and laptops, addressing the growing need for efficient on-device AI solutions. With a focus on optimizing performance without compromising quality, Mirai’s technology includes an inference engine built in Rust that reportedly boosts generation speed by up to 37%. This engine is designed to work seamlessly with minimal code integration, offering developers a streamlined experience similar to Stripe’s payment processing. Mirai’s approach ensures that model weights remain untouched during optimization, preserving output quality while enhancing efficiency. The startup currently focuses on text and voice modalities, with plans to expand into vision in the future. Its framework allows for mixed-mode operations, routing complex requests to the cloud when necessary, ensuring a balance between device capabilities and computational needs. The company’s founders identified a gap in the market: while the conversation around AI often centers on cloud infrastructure, there is a critical need for advancements in on-device AI, especially as consumer hardware continues to evolve. Mirai’s technology not only addresses this gap but also aims to reduce costs associated with cloud inference, aligning with growing concerns over the long-term sustainability of high cloud-computing expenses. Andy McLoughlin of Uncork Capital highlights that the shift toward edge machine learning is inevitable, driven by the rising costs of cloud inference. Mirai’s positioning as a leader in this space makes it well-suited to meet the demand for more efficient AI solutions. By enabling developers to integrate advanced AI features into their applications with

Verticals

techstartups

Originally published on TechCrunch on 2/19/2026