ai-embed-search: A Lightweight Local Semantic Search Engine with Transformer Embeddings and Probabilistic Ranking

We present ai-embed-search, a lightweight, fully local semantic search [8] engine powered by transformer [5] embeddings. Our system supports both local and cloud-based embedding providers, cosine similarity [2] ranking, temperaturescaled softmax scoring, query expansion, and flexible metadata filtering. Unlike traditional solutions relying on external vector databases or APIs, our implementation is designed for fast, memory-efficient, offline usage, making it ideal for constrained or edge environments.