Hamming Distance for Hybrid Search in SQLite
Summary
This article introduces a method to perform semantic search within SQLite by using binary embeddings and the Hamming distance, enabling hybrid search without external vector databases. It explains binary quantization of embeddings, the computation of Hamming distance via XOR and popcount, and shows an example implementation as a SQLite extension with performance benchmarks and practical limits. It also covers how to combine BM25 and semantic results via Reciprocal Rank Fusion (RRF).