Dima Shevts' Mirai raises funds to boost on-device AI inference

The company is developing a framework designed to optimize how AI models perform on consumer devices, with an initial focus on Apple's hardware. According to TechCrunch, Mirai has already built an inference engine specifically for Apple Silicon that is engineered to optimize on-device throughput. This technical approach aims to allow more complex AI tasks to be processed locally on a device's chipset rather than requiring a connection to cloud servers.

Mirai was founded last year by Dima Shevts and Alexey Moiseenkov, both of whom have extensive backgrounds in creating viral, AI-powered consumer applications. Shevts was a co-creator of the popular face-swapping app Reface, which secured backing from the venture firm Andreessen Horowitz (a16z). Moiseenkov was the CEO and co-founder of Prisma, the app that gained widespread popularity last decade for its AI-powered artistic image filters. Their experience gives them a developer-centric perspective on the practical challenges of integrating AI into consumer-facing products.

The founders' interest in on-device AI predates the current generative AI boom. Shevts told TechCrunch that their collaboration began after meeting in London to discuss technology. They identified a significant gap in the market: while the industry conversation was dominated by cloud infrastructure and large-scale data centers, there was a clear need for improved on-device AI capabilities for consumer hardware like phones and laptops. This realization was the catalyst for starting Mirai.

Their hypothesis was confirmed through conversations with other consumer app developers. TechCrunch reports that these developers expressed a desire for better cost optimization and improved margins per token usage, highlighting a clear business need for reducing reliance on expensive cloud-based AI inference. Processing AI workloads on-device can eliminate ongoing cloud computing costs and reduce latency, while also addressing potential privacy concerns by keeping user data local.

The company’s technical work is currently focused on creating a software development kit (SDK). This upcoming SDK will allow other developers to integrate Mirai's optimized runtime into their own applications. The goal is to provide the tools necessary for apps to run sophisticated AI models efficiently on a user's personal device. The $10 million seed funding round, which was led by Uncork Capital, will be used to further this development and expand the company's 14-person technical team.