System Design 75: The Cheatsheet for System Design, LLD and AI-Assisted Interviews
Blind 75 works for coding rounds because it teaches patterns, not questions. Design rounds have patterns too: hot keys, fan-out, real-time delivery, geo lookups, consistency under contention. Learn the pattern once and most "new" questions turn out to be a variation.
This list covers the three design formats you are most likely to meet: system design (HLD), low-level design (LLD), and the newer AI-assisted coding round, where you work in a codebase with an AI assistant and are judged on how well you review what it writes.
How to use this list
- Go in order within a section. Each group starts with the simplest version of its pattern.
- Timebox each question to 45 minutes and follow the same shape every time: requirements, rough numbers, API and data model, high-level design, one deep dive, trade-offs. (See the 45-minute framework.)
- Write down the one concept in the right-hand column before you start. If your answer never touches it, you missed the point of the question.
- 4-week plan: week 1, warm-ups and feeds; week 2, real-time, search and location; week 3, storage, transactions and infrastructure; week 4, LLD and AI-assisted problems, then redo the five you found hardest.
Questions marked free can be practised end to end without paying. For the rest, the first step of each question is free to read on Prepd. AI-assisted problems are part of the AI-Assisted Coding pack.
System design: 40 questions
Warm-ups: counters, presence and short links
| # | Question | The concept to nail |
|---|
| 1 | Design a Like Button (free) | Hot keys: shard the counter, aggregate async |
| 2 | Design a Live Visitor Counter (free) | Approximate counts, TTL heartbeats |
| 3 | Design a Typing Indicator (free) | Ephemeral state, fan-out over WebSockets |
| 4 | Design a URL Shortener (Bitly-like) | ID generation, 100:1 reads, cache |
| 5 | Design a Pastebin / Code Snippet Service | Blob storage vs DB, expiry |
| 6 | Design a Rate Limiter Service | Token bucket, distributed counters |
| 7 | Design a Key-Value Store (DynamoDB-lite) | Partitioning, replication, quorum |
| 8 | Design a Distributed Cache (Memcached-like) | Consistent hashing, eviction, stampedes |
Feeds and fan-out
Real-time and messaging
Search and location
Transactions and consistency
Infrastructure
Low-level design: 10 questions
| # | Question | The concept to nail |
|---|
| 1 | OTP System (free) | Expiry, attempts, rate limits |
| 2 | Parking Lot | Classes, strategy for allocation |
| 3 | LRU Cache | HashMap + doubly linked list |
| 4 | Elevator System | State machine, scheduling |
| 5 | Rate Limiter | Strategy pattern, thread safety |
| 6 | ATM Machine | State pattern, transactions |
| 7 | Flight Booking System | Seat locking, concurrency |
| 8 | Online Auction Platform | Observer, bid ordering |
| 9 | Circuit Breaker Library | States, thresholds, half-open |
| 10 | Access Control System (RBAC) | Roles, permissions, extensibility |
AI-assisted coding: 15 problems
These mirror the AI-assisted round: a small codebase, an assistant, and bugs it will happily introduce. The skill is reviewing its output.
| # | Problem | What the AI tends to get wrong |
|---|
| 1 | Token Bucket Rate Limiter | Refill math, clamping, races |
| 2 | TTL LRU Cache | Expiry + eviction together |
| 3 | Consistent Hash Ring | Virtual nodes, rebalancing |
| 4 | Retrying Payments API Client | Backoff, jitter, idempotency |
| 5 | Wallet Ledger Service | Balances from entries, rounding |
| 6 | Webhook Signature Verifier | HMAC, timing-safe compare, replay |
| 7 | Paginated API Client | Cursors, partial failures |
| 8 | Resumable File Upload | Chunks, checksums, resume |
| 9 | Package Dependency Resolver | Topological sort, cycles |
| 10 | Inventory Hold Service | Holds, expiry, oversell |
| 11 | Live Game Leaderboard | Ranks under concurrent updates |
| 12 | Search Autocomplete | Trie + ranking |
| 13 | LLM Token Chunker | Token limits, overlap, boundaries |
| 14 | Log File Tailer | Rotation, partial lines |
| 15 | CSV Import Validator | Edge cases, error reporting |
What to do after the list
Once the patterns feel familiar, practise saying the trade-offs out loud: what each component fixes, what it costs, and what you would change at 10x the traffic. And when an AI assistant or a teammate hands you a ready-made design, review it the same way. The checklist for reviewing AI-generated code and designs covers what to look for.
Practise it