#deep-dives

Zero-cost architectural decisions — a prompt registry, per-block model config, and three BigQuery columns — that cost nothing to add now and unlock DSPy optimisation, LangSmith experiments, and behavioural flywheels later.

You defined 12 user archetypes. Your data contains 17. How hand-crafted personas fail at tens of millions of users, and what happens when you let BigQuery ML K-Means and diversity-aware sampling tell you the truth.

Cross-modal pattern detection across behavioral data, free-text, and metadata — the capability that actually justifies using an LLM in a recommendation system.