Just brute force your embeddings
SMRTR summary
Many teams building AI search tools assume they need an expensive vector database, but for collections of around 1 million documents with low query traffic, a simple brute-force approach using Python's NumPy library works surprisingly well. A single line of code searching 1 million embeddings returns results in about 12 milliseconds. This saves months of setup time and millions of dollars in infrastructure costs.
SMRTR provides this summary for quick context. The original article belongs to Hacker News.
Read the original article