TBH we are now evaluating ClickHouse in a flow where we need to gather quite a lot of pricing data, and forecast them depending on different dimensions. Based on our calculations we did not want to store that in our main database, and extracted it to a separate one. Hope this decision works out better than the Elasticsearch case I described, but to be honest - we also have projects where it actually made sense, and works until today. Now experimenting with pgvector, and embeddings, that combined with semantic search provide interesting results, especially for flows where we have AI/LLM involved - like MCP search API.
Thanks for sharing the expand-contract pattern case. Glad to hear that!
TBH we are now evaluating ClickHouse in a flow where we need to gather quite a lot of pricing data, and forecast them depending on different dimensions. Based on our calculations we did not want to store that in our main database, and extracted it to a separate one. Hope this decision works out better than the Elasticsearch case I described, but to be honest - we also have projects where it actually made sense, and works until today. Now experimenting with pgvector, and embeddings, that combined with semantic search provide interesting results, especially for flows where we have AI/LLM involved - like MCP search API.
Thanks for sharing the expand-contract pattern case. Glad to hear that!