Roman Shemet is Co-Founder and CEO of Cactus Compute (YC S25), an open-source on-device AI inference engine built to run large language models directly on smartphones, wearables, and edge hardware. He is also a member of the Forbes Business Council and a vocal advocate for the future of local, private mobile AI.

Roman holds an MSc in Financial Economics from the University of Oxford and built his career as a quantitative researcher and economist, with additional expertise in product design and data engineering. He previously worked in equity investing and algorithmic trading before pivoting to build AI infrastructure.

Roman co-founded Cactus after connecting with co-founder Henry Ndubuaku through YC co-founder matching in London, united by the belief that the future of AI belongs on-device rather than in the cloud. Outside of work, he is a licensed pilot, competitive triathlete, and chess enthusiast.

Cactus Compute is a Y Combinator S25 company backed by Antigravity Capital and other investors, with Roman leading the product and go-to-market strategy as CEO. The company has grown to an eight-person team in San Francisco and hosted a Mobile Agent Hackathon in late 2025 to grow its developer ecosystem.

Cactus already powers over 500,000 weekly inference tasks in production, achieves latency under 100ms on flagship devices, and has amassed 4,200+ GitHub stars as a fully open-source project. The platform supports any GGUF model from HuggingFace and offers SDKs for Flutter, React Native, Kotlin, and Swift.