ZORN SIGHT — Fast Multimodal Edge AI

ZORN SIGHT is a fast, offline-first multimodal AI system designed to bring capable AI directly to edge devices. Instead of relying on cloud APIs, ZORN SIGHT runs locally and combines text chat, image understanding, and video understanding in one compact system. The project is built around SmolVLM2 and focuses on the inference layer rather than simply running a pretrained model. ZORN SIGHT introduces a hardware-aware runtime architecture with optimized preprocessing, smart video frame sampling, memory-aware inference, and Intel CPU acceleration through OpenVINO when available. For video understanding, the system selects relevant frames while preserving frame indices and timestamps, allowing responses to be connected to visual evidence. The architecture also supports backend selection and hardware detection, with PyTorch as a general fallback and an extensible path toward Qualcomm QNN deployment. ZORN SIGHT is designed for local and offline operation after the initial model download. This makes multimodal AI more practical for environments where cloud connectivity, privacy, latency, or compute resources are constraints. The goal is simple: make multimodal AI faster, more efficient, and deployable at the edge across different hardware targets.