CreativAI Builds the ‘SQL Layer’ for Physical AI, Turning Raw Video into Queryable Databases
CreativAI is building the infrastructure to transform raw camera footage into a structured, queryable database. Founded by former Meta FAIR researcher Mohamed Elhoseiny and Waleed Shaarani, the company is engineering what the founders call the "SQL layer for Visual and Physical AI."
The startup targets a structural limitation in modern computer vision. While artificial intelligence can effectively categorize images, enterprises struggle to build scalable applications on top of raw visual perception. CreativAI replaces unstructured video files with a continuous knowledge base, allowing autonomous agents, robots, and human operators to query physical environments exactly as developers query structured data.
Moving Beyond Larger Models
During his tenure at KAUST, Stanford, Adobe, and Meta FAIR, Elhoseiny observed that scaling general-purpose AI models fails to solve enterprise-specific visual challenges. Businesses operate in a long tail of rare, highly specific operational events that general training datasets rarely capture.
"The question is no longer whether a machine can understand an image. The question is what humans can actually do with that understanding at scale," Elhoseiny notes, defining the catalyst for CreativAI.
Rather than training new bespoke models for every enterprise question, CreativAI processes video into structured entities, events, and behaviors. This architecture delivers two distinct technical advantages:
- Traceability: The system links every structured record directly back to the exact video frame, providing verifiable evidence for high-assurance environments.
- Shared Memory: Robots, AI agents, and human operators access a single, continuously updated record of reality, eliminating the need for fragmented perception systems.
Vertical Sequencing and Enterprise Deployment
The platform holds immediate utility for robotics and autonomous vehicles, which require persistent memory to reason over their surroundings. The architecture also maps to airports, factories, hospitals, and logistics networks—environments equipped with thousands of disconnected cameras that lack shared operational intelligence.
To manage this broad applicability, the founders apply strict vertical sequencing to their go-to-market strategy. Instead of pursuing multiple industries simultaneously, CreativAI targets a single opportunity, deploys the core engine, and wins the vertical before adapting the same underlying infrastructure for the next market.
The startup currently works directly with enterprise design partners to bypass self-service API limitations and refine the product in live operational environments. Major technology providers back the effort, with CreativAI participating in startup accelerator programs from AWS, Google for Startups, Microsoft for Startups, and NVIDIA Inception. These partnerships supply the heavy compute resources required for AI infrastructure development.
Infrastructure for Autonomous Systems
As capital accelerates the deployment of physical robots and autonomous vehicles, the volume of raw visual data expands. Autonomous machines require structured information to function, pushing visual intelligence from a standalone analytics feature to a core infrastructure requirement.
CreativAI faces the immediate task of transitioning its research-driven architecture into reliable enterprise deployments. The company must prove its structured data approach can scale across industrial environments, establishing its platform as standard tooling for organizations operating intelligent machines.