Vision AI

Technical Description

• Vision AI is a high-performance, privacy-first mobile application built with React Native and Expo that enables semantic image search using fully on-device AI.
• Semantic Search: Search for images using natural language (e.g., "sunset at the beach" or "dog playing in the park") rather than just filenames or tags.
• 100% On-Device Inference: Utilizes 8-bit quantized ONNX models for both vision and text embeddings, ensuring maximum privacy and offline capability.
• Vector Search Engine: Powered by sqlite-vec, performing high-speed cosine similarity searches across thousands of locally stored image embeddings.
• Optimized Performance: Leverages ONNX Runtime and custom asset interceptors to minimize latency and memory overhead on mobile hardware.
• Implemented a Custom Fetch Interceptor for @xenova/transformers that redirects model/tokenizer requests to the local Expo asset system, enabling a strictly air-gapped search experience.

Tech Stack