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

Source: TechCrunch

Much of the conversation around AI today is focused on building cloud capacity and massive data centers to run models. Companies like Apple and Qualcomm are in the early stages of making on‑device AI more useful. Amid all that, the 14‑person technical team of London‑based Mirai is working to improve how models run on phones and laptops.

Founders and Origin

Mirai was founded last year by Dima Shvets and Alexey Moiseenkov, both seasoned consumer‑app builders.

• Dima Shvets co‑founded the face‑swapping app Reface, which was backed by a16z. He later became a scout for the venture firm.
• Alexey Moiseenkov was CEO and co‑founder of the viral AI‑filters app Prisma.

Both founders had been thinking about AI and machine learning on devices long before generative AI became popular.

“When we met together in London, we started to chat about technology, and we realized that within the hype of gen AI and more AI adoption, everybody talks about cloud, about servers, about AGI coming. But the missing piece is on‑device AI for consumer hardware,” Shvets told TechCrunch.

Their desire to enable complex tasks on the phone led them to start Mirai. When they surveyed other consumer‑app developers, many expressed a need for better cost optimization and higher margin per token usage.

Technology and Product

Mirai is developing a framework that makes models perform better on devices. Key points:

• Inference engine for Apple Silicon – optimizes on‑device throughput.
• Upcoming SDK – lets developers integrate the runtime with just a few lines of code.
• Rust implementation – claims up to a 37 % increase in generation speed without altering model weights, preserving output quality.
• Current focus – improving text and voice modalities; vision support is planned.
• Edge‑cloud orchestration – an orchestration layer will route requests that cannot be satisfied on‑device to the cloud, enabling mixed‑mode operation.
• Future platforms – plans to bring the engine to Android and to release on‑device benchmarks for model makers.

The engine could eventually power on‑device assistants, transcribers, translators, and chat applications.

Funding and Investors

Mirai closed a $10 million seed round led by Uncork Capital. Participants also included a range of individual investors:

• David Singleton (Dreamer CEO)
• François Chaubard (YC Partner)
• Marcin Żukowski (Snowflake co‑founder)
• Staniszewski (ElevenLabs co‑founder)
• Gokul Rajaram (former Google AdSense product manager, Coinbase board member)
• Scooter Braun (Groq investor)
• Vijay Krishnan (Turing.com CTO)
• Ben Parr and Matt Schlicht (Theory Forge Ventures)
• Aditya Jami (ex‑Netflix technical leader)

Andy McLoughlin, managing partner at Uncork Capital, noted that while cloud inference costs are currently high, the economics will eventually push more workloads to the edge. He believes Mirai is well positioned to capture that demand.

Future Outlook

Mirai’s roadmap includes:

• Expanding support to vision models.
• Launching Android compatibility.
• Providing on‑device benchmark tools for the broader AI community.
• Continuing collaborations with frontier model providers and chipmakers to fine‑tune models for edge deployment.

The company aims to give developers a “Stripe‑like” experience—just a few lines of code to integrate capabilities such as summarization, classification, or other AI use cases—while keeping inference efficient and cost‑effective on consumer hardware.