learn.smmeyer.devdrills
menu
practice / drills
guided practice
Apply one course concept at a time. Choose your track, make an attempt, compare the worked answer, and revise the weakest part. Start with either exercise after its preparation chapters. Use it for part of that chapter's practice budget.
Each case supplies its facts and expected artifact. Calculation checks are automatic; explicit rubrics help you assess the reasoning. Your first attempt and revision stay together in local progress, so you can resume or repeat the same case later.
01 / System Design · ~20 min
Contain a cache stampede
Calculate source pressure, trace concurrent expiry, and design a refresh path that stays safe when the cache or refresh owner fails.
Produce a worksheet, compare a worked answer, and save a specific revision.
Useful after
Start guided exercise →02 / LLM Engineering · ~20 min
Repair an unsupported policy answer
Separate authorized current evidence, context-assembly failure, and unsupported generation; produce a supported answer and a safe missing-evidence response.
Produce a worksheet, compare a worked answer, and save a specific revision.
Useful after
Start guided exercise →
Independent interview practice · scenario deck, redo queue, and oral fluency
After guided practice, use the original random deck for open-ended design rehearsal. These prompts require independent self-review. For a timed full design, use the 50-minute mock workspace →
Draw to get one scenario plus seven constraint cards. Same seed replays the exact draw — redo it after 48 hours without reviewing the old diagram.
48-hour redo queue
system design
→ not yet attempted
system design
→ not yet attempted
system design
→ not yet attempted
system design
→ not yet attempted
system design
→ not yet attempted
system design
→ not yet attempted
system design
→ not yet attempted
system design
→ not yet attempted
system design
→ not yet attempted
system design
→ not yet attempted
AI-native design
→ not yet attempted
AI-native design
→ not yet attempted
AI-native design
→ not yet attempted
AI-native design
→ not yet attempted
AI-native design
→ not yet attempted
AI-native design
→ not yet attempted
AI-native design
→ not yet attempted
AI-native design
→ not yet attempted
AI-native design
→ not yet attempted
AI-native design
→ not yet attempted
constraint decks
scale · 6 cards
- · 10k DAU side project, one engineer
- · 20M MAU consumer app
- · 1B events per day pipeline
- · 100k RPS flash-sale spike
- · 10-person startup, two backend engineers
- · five regions on day one
traffic shape · 6 cards
- · steady diurnal curve
- · spiky: 10× during live events
- · write-heavy 9:1 ingest
- · read-heavy 100:1
- · long tail: 1% creators, 99% lurkers
- · batch bulk load plus interactive queries
SLO · 6 cards
- · 99.9% monthly, 500ms p99
- · 99.99% checkout, 200ms p99
- · best-effort internal tool
- · 4-hour RPO, 15-minute RTO
- · zero data loss; minutes of downtime tolerable
- · realtime: 100ms glass-to-glass
failure · 6 cards
- · primary region lost mid-session
- · cache fleet cold-restarts at peak
- · noisy tenant eats 70% of GPU capacity
- · leader election storms every hour
- · core dependency p99 doubles overnight
- · provider bills 10× after a config typo
security / abuse · 6 cards
- · public API with active scraping
- · multi-tenant PII under SOC 2
- · prompt-injection attempts inside every upload
- · DDoS during launch week
- · compromised third-party webhook
- · credential-stuffing against login
data residency · 6 cards
- · EU data stays in the EU
- · single region to start
- · US plus EU with in-region inference
- · customer-managed keys per tenant
- · no PII leaves the device
- · must stay reachable from mainland China
cost · 6 cards
- · ramen budget: $500/mo infra
- · GPU spend capped — queue the rest
- · optimize for spend per successful task
- · egress costs dominate the bill
- · scale to zero every night
- · unit economics must beat $0.01/user/mo
oral fluency circuit
Out loud, timed, no pauses. Seniority is audible: short sentences, named tradeoffs, explicit risks.
oral fluency · 2 min
Whiteboard an LLM for a backend engineer, then name one product consequence.
2:00