Classifying your workload first. Workload type: OLTP — user-facing, mixed read/write, low latency required. Data shape: Highly relational. Posts belong to users, likes belong to posts and users, follows are a many-to-many relationship between users, comments nest under posts. Recommendation: PostgreSQL. MongoDB's document model gives you a locality advantage when one root entity owns everything beneath it — a résumé, a product listing, a user profile. Your social app is not that. You have entities that reference each other in every direction. Storing that in MongoDB means either deeply nested documents that become impossible to query across, or storing IDs and doing manual joins in application code — which is slower, more error-prone, and gives up the one thing MongoDB is actually good at. PostgreSQL handles this with proper foreign keys, join indexes, and referential integrity enforced at the database level. Add Redis in front for feed caching and session storage once you need it. Stack: PostgreSQL as your primary store, Redis for caching hot feeds and sessions