PyOhio 2026
Vector databases are the hot new infrastructure tax — Pinecone, Weaviate, and managed OpenSearch all want $70-300+/month before you've stored a single embedding. In this talk, I'll show how we replaced a planned vector database with AWS S3 Vectors — a new S3 feature that stores and queries embeddings directly in S3 buckets — to power semantic search across millions of genealogical records using Python and Cohere embeddings via AWS Bedrock, all for under $1/month.
I'll cover how S3 Vectors works under the hood, walk through a live Python implementation (embedding generation, indexing pipeline, and query interface), share real production cost comparisons against OpenSearch Serverless and Pinecone, and discuss when S3 Vectors is the right choice vs. a dedicated vector DB. This talk is for Python developers who want to add semantic search to their applications without adding another expensive managed service to their stack.
