Writing
Long-form notes on production AI systems and architecture: agentic workflows, retrieval, evaluation, observability, and the control layer around real products. Mix of essays and system-design documents.
-
Engine vs Bolt-on: where AI in online learning actually stands
The real shift in AI for online learning isn't about content or motivation — it's about the control loop. Where AI sits relative to the learning runtime is the split that matters.
-
AI-Native Learning System — Adaptive Lesson Runtime
System design for a real-time adaptive teaching system that delivers personalized learning from structured ingredients.
-
Content Ingestion — Structured Learning Ingredient Pipeline
Upstream system that converts expert knowledge and course assets into structured, runtime-ready learning ingredients.
-
AgentOps — reliability, evals, tracing
Tool scopes, permissions, audit trails, regression gates, and tracing for agentic workflows in production.
-
On building measurable LLM systems
Why reliability, cost, and latency matter in production LLM systems — and how to align architecture and evals with product goals.