Agentic AI referee that watches video, reasons with chain-of-thought, and signals physical robots. Multi-provider VLM switching with a full agent console showing the Observe-Perceive-Reason-Decide-Act loop.
View Live Project → GitHub →Sideline is a physical AI system that bridges the gap between digital reasoning and real-world action. It watches sports video through vision language models, reasons about what it sees using chain-of-thought analysis, and then signals physical robots to act on its decisions.
The full agent console exposes every step of the cognitive loop — Observe, Perceive, Reason, Decide, Act — so you can see exactly how the AI arrives at each call. Multi-provider VLM switching means you can swap inference backends without changing the pipeline.
Qwen VLM handles vision-language reasoning, Nebius provides scalable inference, and Jetson AGX Orin runs the edge pipeline. Three.js powers the 3D robot visualization, with WebSocket for real-time communication between the AI agent and physical hardware.
We build AI agents that reason, decide, and act in the physical world.
Book a Free Call →