Teaching a child in <1000 ms: the architecture behind a real-time tutor
Summary
The article details building a real-time AI tutor for children aged 4-9, focusing on sub-second, pedagogically grounded interactions. It describes a streaming harness that separates generation from execution (converser, planner) to reduce latency, plus async planning, immutable event logs, and a safety classifier that gates actions without delaying generation. The piece also discusses safety, pedagogy, and broader challenges of engineering a real-time, child-facing AI tutor.