I write about how LLMs work under the hood, software architecture patterns, and the craft of building reliable systems. Topics include tokenization, attention, constrained decoding, tool use, and agentic architectures.