You are designing the data layer for an AI engineering platform with three different storage needs: (1) user profiles, billing, and subscription data, (2) 50M embeddings for semantic search across documents, (3) conversation history for 100k daily active users with fast recent-message retrieval. For each requirement, select and justify your database choice. Then design how these three databases interact in a single request flow. When would you add a graph database? When would a document store outperform a relational DB for AI workloads?