Notes on applied AI, retrieval-augmented generation, and shipping LLM products.
ADK 2.0 ships a three-layer skill loading pattern that cuts baseline context by around 90%. Architecture, four skill patterns, and the code you'd actually write to adopt it.
