Andrey Mikhaylov is the creator of TurboFieldfare, a groundbreaking open-source local inference engine built in Swift and Metal that runs Google's Gemma 4 26B-A4B model in approximately 2 GB of RAM on Apple Silicon Macs.
Andrey is a Senior iOS and Metal Engineer currently at Lapse and previously at Prisma AI, based in London, UK, with deep expertise in images, video, and on-device AI.
Andrey engineered TurboFieldfare from scratch using Apple's Metal GPU framework and Swift, conducting over 100 optimization experiments to achieve its unprecedented memory efficiency.
TurboFieldfare hit #1 on Hacker News on July 29, 2026 with 218 points, and delivers 5-6 tokens per second on an M2 Air and 31-35 tokens per second on an M5 Pro.
The project democratizes access to large language models by enabling a 26-billion-parameter model to run on an 8 GB MacBook Air, offering a native app, CLI, and an OpenAI-compatible local server.
3309
Andrey Mikhaylov
Company(s)
Industry
SaaS
other founders
