New series · 8 episodes
The ML production lifecycle — the part you never actually build
Data labeling → eval harness → training → finetuning → inference & serving → monitoring. One stage per episode, on systems we actually ship. Whether you’re shipping it in prod or nailing the ML-system-design interview, it’s the same hands-on work — and almost nobody teaches it hands-on.
What you’ll get
You run the pipeline
Every episode ships a runnable repo — sanitized from real production code, not a toy notebook. You run the eval harness, reproduce the skew, prove the finetune.
Real systems, not slides
One lifecycle stage per episode, anchored to a system we actually ship. The gotcha, the war story, the thing the tutorial-of-record leaves out.
The round interviews now test
2026 ML interviews grade eval methodology, MLOps fluency, and production debugging — not leetcode speed. This is the hands-on version incumbents only whiteboard.
We publish the research and engineering behind the apps we ship. This series is that work, in the open — free to watch, with the repo to run yourself.