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

Co-founders Dima Shvets and Alexey Moiseenkov of Mirai are revolutionizing on-device AI modeling to improve functionality in consumer hardware. Their startup, backed by a $10 million seed investment, is focused on creating an inference engine optimized for Apple Silicon, with plans to release a developer-friendly SDK. This innovation addresses the growing need for efficient AI processes without heavy reliance on cloud computing.

Key Points

• Mirai aims to improve on-device AI model inference for consumer hardware (phones, laptops).
• Founded by Dima Shvets (Reface co-founder) and Alexey Moiseenkov (Prisma co-founder) to tackle challenges in on-device AI.
• The startup raised $10 million in seed funding led by Uncork Capital.
• Developed an inference engine for Apple Silicon, enhancing model throughput.
• Upcoming SDK will allow developers to integrate models easily with minimal code.
• Built in Rust for speed improvement (up to 37%), without compromising model quality.
• Currently focusing on text and voice modalities, with future plans for vision support.
• Developing benchmarks for on-device performance and an orchestration layer for cloud integration.

Relevance

• The trend towards on-device solutions responds to increasing cloud inference costs and growing consumer demand for faster, efficient AI.
• Historically, reliance on cloud computing has driven up operational costs, prompting a shift towards edge computing.
• As generative AI gains traction, emphasis is shifting from cloud-based to on-device capabilities, aligning with the 2025 IT trend of decentralized AI processing.

Mirai's innovative approach to on-device AI modeling addresses critical market needs for efficiency and cost-effectiveness, positioning the company at the forefront of emerging technology trends as it anticipates a transition towards edge computing solutions.