Fast filtered vector search (without a database engine)

Today many AI/ML systems like recommenders, image search, or text search rely on finding the most similar vectors for a given vector. As exact brute force search is linear in the number of vectors to search in, brute force search is too slow for systems with millions of vectors…

When hash-maps are an order of magnitude too slow

Recently I was writing some performance critical code. It was about an algorithm which had to run in linear time. Out of habit, I started to use hash-maps in order to make linear time possible. The execution speed was not as expected so I profiled the code. Hash-map lookups were…