Notes on architecture, AI/LLMs, agentic systems and the engineering craft.
Beyond the demo — what it actually takes to run an LLM-powered feature in production: evaluation, cost control, latency, and failure modes.
