NoSQL data modeling inverts relational thinking:
Query-driven design: Model for your access patterns, not for data relationships. Duplicate data to avoid joins.
Embedding vs referencing: Embed related data if always accessed together. Reference if data is large or independently accessed.
Example: Blog with comments. SQL: posts and comments tables with foreign key. NoSQL: embed comments in post document. Faster reads, but updating a comment requires rewriting the post.
Denormalization trades storage for read performance. Good for read-heavy workloads.