Ahmed Ahres

Ahmed Ahres is a computer-vision engineer, startup founder, and product-and-go-to-market leader at Reactor, where he is helping developers build applications around real-time interactive video. He previously co-founded Nativ, an Andreessen Horowitz-backed startup that adapted visual content for different languages and cultures.

Raised in Tunisia, Ahres was a national tennis champion before studying computer science in the Netherlands and shipping mobile games. His early projects included visual-marker detection with OpenCV and satellite-image classification using convolutional neural networks.

He joined Revolut as its first technical intern, building facial-authentication and fraud-prevention systems before becoming a full-time machine-learning engineer. As a founding engineer at Heron Data, he expanded into product development, sales, and customer relationships. He subsequently pursued entrepreneurship through Entrepreneur First and worked as a senior machine-learning engineer at augmented-reality company XYZ Reality.

In 2024, Ahres co-founded Nativ with former Revolut colleague Ashish Ranjan Jha and became its chief technology officer. The company focused on AI-powered visual culturalization: modifying images, video, and marketing materials to reflect the languages and cultural expectations of different audiences. Nativ participated in Andreessen Horowitz’s Speedrun accelerator; in September 2025, Ahres announced that he had become a Speedrun scout.

After leaving Nativ, he joined Reactor and began hosting Behind the Worlds, a podcast about creators working across games, film, animation, visual effects, and immersive environments.

• Video should behave like software. Ahres treats world models as continuously generated environments that users can redirect while events unfold. Immediate feedback gives creators and audiences control unavailable in conventional systems that deliver completed video files.

• Interactive worlds can become educational environments. His proposed applications extend beyond games to audience-directed livestreams, immersive history lessons, medical training, cooking simulations, and prompt-driven video editing.

• Streaming world-model infrastructure is its own engineering problem. Responsive applications require continuous pixel delivery, persistent sessions, contextual memory, geographically distributed GPUs, and low latency. Reactor makes third-party models available to developers; Ahres does not claim to have created those underlying models.

• Consistency and evaluation remain unresolved. Generated worlds can forget previous details, avatars still feel unnatural, and dependable measurements of temporal consistency remain immature. In his AI Engineer World’s Fair talk, Ahres identified multi-GPU execution, model-weight optimization, and quantization as potential ways to improve responsiveness.