qdrant/qdrant-edge-demo

Embeds video frames as vectors using CLIP ViT-B-32, storing them locally in Qdrant Edge's dual-shard architecture (mutable for real-time search, immutable for indexed snapshots) while a SQLite queue syncs data to a remote Qdrant server for HNSW indexing. The system performs text-to-vector search with maximal marginal relevance (MMR) queries across both shards, enabling offline visual search with eventual consistency through incremental snapshot syncing that avoids full database downloads.