[AI Summary]: This blog post by Leonie Monigatti shares 37 key insights gained from two years working at Weaviate, a vector database company. The comprehensive list covers fundamental concepts in information retrieval, from the importance of BM25 as a baseline search method to advanced topics like vector quantization and hybrid search strategies. Key takeaways include understanding that vector search is approximate rather than exact, the distinction between different types of embeddings (dense, sparse, binary), the economics of vector dimensions, and the critical differences between similarity and relevance. The post emphasizes practical considerations for implementing RAG pipelines, including chunk size optimization, the role of tokenizers, and when to use keyword versus vector search. Monigatti also addresses common misconceptions about vector databases and highlights that while vector search is powerful, it’s just one tool in the retrieval toolbox that should be combined with filtering, reranking, and other techniques for optimal results.