Existing semantic search solutions, which rely on OpenAI's embedding model and vector databases like Pinecone and Chroma are expensive. After running into this issue multiple times, we've built a cheaper and simpler alternative using Instructor embeddings and Elasticsearch.