Nhung Luong

Notes on data engineering, AI engineering, and building data products.

Pipelines, models, and the messy bits in between — notes on building data platforms that hold up in production, from ingestion and warehousing to the AI agents running on top of them, plus whatever breaks along the way.

Aug 6, 2026 · 10 min read

Walking through prompt caching, turn by turn

A three-turn worked example of Claude's prompt caching — what gets read, what gets written, where the 20-block lookback quietly breaks it, and what it's actually worth in dollars.

Jul 30, 2026 · 6 min read

Same prompt, better slide: why LLMs draw nicer HTML than PPTX

Why the same prompt produces a better-looking HTML/SVG slide than a .pptx file, with a worked example and the two-step workflow that borrows the win.

Jul 29, 2026 · 4 min read

What Tesla's delivery speed actually teaches: decouple the platform from the feature

Notes on an analysis contrasting Tesla and German automakers — and why the real lesson is about software platforms, not cars.

Jul 26, 2026 · 9 min read

IBM Industry Models for Banking

A layer-by-layer walkthrough of IBM Industry Models for Banking — conceptual, logical, physical — with a worked example and diagrams.

Jul 23, 2026 · 2 min read

The AI layoff trap: when cutting headcount backfires

Notes on a paper arguing that AI-driven layoffs can be individually rational but collectively self-defeating.

Jul 21, 2026 · 2 min read

The one habit that saved my weekends — idempotent pipelines

Why a pipeline you can safely re-run is worth more than a clever one, and how to build it.