Relling

Relling is working on the “ImageNet for world models”: large-scale, high-quality multimodal datasets that help AI and robotics teams train systems to see, move, and interact with the real world. We combine synchronized video, depth, LiDAR, motion, and audio, we give researchers the building blocks for truly general-purpose AI.

Relling is a startup building large-scale, high-quality multimodal datasets—described as the “ImageNet for world models”—to enable AI and robotics systems to better perceive, move, and interact with the real world. Their product combines synchronized video, depth, LiDAR, motion, and audio data, providing foundational infrastructure for training general-purpose AI models that understand complex, real-world environments. This serves AI researchers and robotics teams who need rich, multimodal data to develop more robust and capable systems, addressing the limitations of text- or image-only models. Relling is in early growth stages, having recently launched and begun hiring, with a focus on enabling the next generation of AI trained on raw video experience rather than just text or images[1][2][3].

Founded in 2025 by Jai Relan and Anya Singh, Relling emerged from the recognition that current AI models trained on text or images alone are insufficient for capturing the full complexity of real-world interactions. Jai Relan brings experience as a serial founder, venture capitalist, and product leader at notable startups, while Anya Singh adds a unique perspective as a chess player, likely contributing strategic thinking. The company is based in San Francisco and was part of Y Combinator’s Summer 2025 batch, marking a pivotal moment in gaining early traction and visibility within the AI and robotics ecosystem[2][3].

Core Differentiators

• Unique Data Infrastructure: Relling provides synchronized multimodal datasets (video, depth, LiDAR, motion, audio) that capture real-world events in high fidelity, enabling AI models to learn from raw video experience rather than engineered features or single modalities[1][3].
• Focus on World Models: Unlike traditional datasets focused on text or static images, Relling’s data supports training AI systems that can see, predict, and act in dynamic environments, closing the loop between perception and action[3].
• Developer and Researcher Enablement: By offering large-scale, high-quality data, Relling reduces the engineering overhead for AI teams, accelerating development cycles and improving model robustness to occlusion, blur, and complex 3D interactions[1][3].
• Early-Stage but High Potential: As a YC-backed startup with experienced founders, Relling benefits from strong network access and mentorship, positioning it well to influence the future of AI infrastructure[2].

Role in the Broader Tech Landscape

Relling is riding the wave of AI’s shift from text- and image-based models to *multimodal world models* that integrate diverse sensory inputs to better understand and interact with the physical world. This trend is driven by the growing demand for AI in robotics, autonomous systems, and augmented reality, where understanding 3D space and temporal dynamics is critical. The timing is crucial as advances in sensors (LiDAR, depth cameras) and compute power now make it feasible to collect and process such rich datasets at scale. Relling’s infrastructure addresses a key bottleneck—high-quality, synchronized multimodal data—thus enabling breakthroughs in general-purpose AI and robotics[1][3].

By providing foundational datasets, Relling influences the broader ecosystem by lowering barriers for startups and research labs working on embodied AI, robotics, and real-world perception. Their work could accelerate innovation across industries reliant on AI that understands complex environments, from autonomous vehicles to smart manufacturing.

Quick Take & Future Outlook

Looking ahead, Relling is poised to become a critical infrastructure provider for the next generation of AI models that require rich, multimodal sensory data. As AI systems increasingly move beyond narrow tasks to general-purpose world understanding, demand for Relling’s datasets and tools will likely grow. Trends such as robotics automation, augmented reality, and AI-driven simulation will shape their journey, pushing them to expand dataset scale, diversity, and accessibility.

Their influence may evolve from a data provider to a platform enabling entire ecosystems of AI and robotics innovation, potentially integrating with model training frameworks and developer tools. Success will depend on their ability to maintain data quality, scale operations, and foster a strong community of users who rely on their infrastructure to build the future of AI.

In summary, Relling’s mission to create the “ImageNet for world models” situates them at the forefront of a pivotal shift in AI, providing essential data infrastructure that could redefine how machines perceive and act in the real world[1][2][3].