Revision 10 — Recall & Drill: Data Modelling + LLM Serving
Spaced-repetition recall + drills for the two prior sessions.
Why revision matters
Spaced repetition is what turns understanding into retention. Every third session of this series is a revision block — no new content, just active recall and hands-on drills for the two topics you just covered.
If you're consuming this cold, treat these revisions as self-assessment. If you can't hit the recall prompts, go back and re-read the linked posts before moving on.
The two sessions you're revisiting
- S19 — Data Modelling — Dimensional, Data Vault, One Big Table
- S20 — LLM Serving — KV Cache, Batching, Speculative Decoding
Session structure (90 min)
| Time | Activity |
|---|---|
| 0–10 min | Warm-up flashcards — 10 rapid-fire recall prompts |
| 10–35 min | Blank-page recall — write everything you remember, no notes |
| 35–60 min | Redo the hands-on — one code snippet or diagram, from scratch |
| 60–80 min | Redo one LeetCode / system-design drill from either session |
| 80–90 min | Gap analysis — list what you missed, plan a re-read window |
Warm-up flashcards (10)
Prompts you should be able to answer in under 30 seconds each. Cover the answer, say it out loud, then check.
- What was the one-line motivation for the first session's topic?
- What was the one-line motivation for the second session's topic?
- Draw the core diagram from S19 from memory — what are the 3–5 boxes and their arrows?
- Draw the core diagram from S20 from memory — what are the 3–5 boxes and their arrows?
- State one number (throughput, latency, cost, ratio) from either session and what it constrained.
- What is one failure mode either topic guards against?
- What is one anti-pattern either session called out?
- If you had to teach either topic to a colleague in 3 minutes, which concept would you lead with?
- Which primitive or abstraction is shared between the two topics?
- What is one thing you still don't understand — write it down, look it up this week.
Blank-page recall (25 min)
Take a blank page or scratch file. For each of the two sessions:
- Write the topic in the middle.
- Branch out with every concept, term, formula, code snippet, and system component you remember.
- Draw the core diagram from memory.
- Don't peek at the original post. Whatever you write down is what you actually own.
When done, open the linked posts side-by-side and colour-code:
- ✅ Green — you had it
- 🟡 Yellow — you had it but fuzzy
- 🔴 Red — you missed it entirely
The reds are your revision debt for this week.
Redo the hands-on (25 min)
Pick one hands-on element from either session:
- A code snippet — reproduce it from scratch in a fresh file. No copy-paste. No looking at the original until you're stuck for 5 minutes.
- A diagram — redraw it on paper or tldraw, then compare to the original.
- A calculation — recompute one of the "typical numbers" tables and explain each cell.
If it takes more than 25 minutes, you don't own the material yet. Note the gap, move on, revisit next weekend.
Redo one drill (20 min)
Pick one:
- The LeetCode problem from either session — redo it without hints. Log your time. If you were faster than last time, note the speedup.
- The system-design prompt — draw the high-level architecture on a blank sheet in 20 minutes flat.
Gap analysis (10 min)
Write in your prep journal:
- Three things I got right without effort.
- Three things I fumbled or forgot.
- One concept I'd re-read this weekend.
- One drill I'd redo before the next revision.
Post-session checklist
- Flashcard round complete — record the score / 10.
- Blank-page recall done — red/yellow gaps logged.
- One hands-on redone from scratch.
- One drill re-solved with timer.
- Gap-analysis entry in prep journal.
- Committed notes to your prep repo with message
revision-10: <one-line>.
No new content. No new tabs. Just recall, drill, and log. That's the whole session.