Skip to content

04. Week 17 — Daily Recall

Use these cold. Then check the cited sections in 02_explainer.md or 03_study_material.md.

Monday — Opening failure

  1. Why is a notebook win not a production win? See 02_explainer.md §1.1-§1.8.
  2. What made the three-week degradation silent? See 02_explainer.md §1.3.
  3. What operational questions does a senior engineer ask immediately? See 02_explainer.md §1.6.

Tuesday — Lifecycle management

  1. What must every tracked run contain? See 02_explainer.md §2.2-§2.3.
  2. Why is a model registry more than a folder? See 02_explainer.md §2.5-§2.6.
  3. Which four things must be versioned together in ML? See 02_explainer.md §2.7.
  4. Why is reproducibility a stack, not a wish? See 02_explainer.md §2.8-§2.10.

Wednesday — CI/CD for ML

  1. What does the quality gate actually verify? See 02_explainer.md §3.5.
  2. Champion versus challenger — why use that language? See 02_explainer.md §3.6.
  3. When is automated retraining wise, and when is it reckless? See 02_explainer.md §3.7.
  4. When does a feature store help, and when is it unnecessary? See 02_explainer.md §3.8-§3.9.

Thursday — Serving infrastructure

  1. vLLM vs TGI vs Triton — when would you choose each? See 02_explainer.md §4.3 and 03_study_material.md §4.
  2. Why is QPS alone a weak autoscaling signal for LLMs? See 02_explainer.md §4.4.
  3. Static, dynamic, and continuous batching — what problem does each solve? See 02_explainer.md §4.5.
  4. What are the first cost levers before quantization? See 02_explainer.md §4.8.

Friday — Monitoring and maintenance

  1. Data drift versus model drift — explain the difference cleanly. See 02_explainer.md §5.3-§5.5.
  2. What makes A/B testing harder for AI systems? See 02_explainer.md §5.7.
  3. Shadow versus canary versus blue-green — what does each catch? See 02_explainer.md §4.10 and §5.8.
  4. What exactly can you roll back in an AI system? See 02_explainer.md §5.9.
  5. What belongs in an AI incident runbook? See 02_explainer.md §5.10.

Weekend synthesis

  1. Explain the full factory analogy in two minutes. See 02_explainer.md §ELI5.
  2. Draw the lifecycle from memory: run → warehouse → assembly line → quality gate → production monitor.
  3. Give one concrete tool stack for a small team and one for a cloud-heavy team. See 02_explainer.md §6.4.
  4. Say the bridge sentence into Module 18 without notes. See 02_explainer.md §What Comes Next.