smartscan-lib

smartscan-lib is a Python library that provides core ML and data infrastructure for on-device AI applications. According to the README, it powers the SmartScan Server used by the Desktop App. The library exposes quantized embedding providers. For images: CLIP ViT-B-32, DINOv2 Small, and Inception ResNet V2 for facial recognition. For text: CLIP ViT-B-32, all-MiniLM-L6-v2, and all-distilroberta-v1. A ModelManager gives access to image and text embedders, which support single and batch embedding. Indexers are built on a BatchProcessor abstraction and include ImageIndexer, VideoIndexer, and DocIndexer. They accept URLs or local paths, use the configured embedder, and can take an optional ProcessorListener for progress and batch callbacks. Additional components include IncrementalClusterer for grouping embeddings by similarity with adaptive thresholds, existing assignments and clusters, and optional auto-merging; and few-shot classification via few_shot_classify or FewShotClassifier, comparing embeddings against pre-labelled cluster centroids with batch support. Installation requires Python 3.10+ and can be done from the Git repository with pip. The README focuses on library usage examples rather than deployment details.