CODING PATTERNS9 min read · 2026-11-20
Stop memorizing 1,000 disconnected LeetCode solutions. Use input constraints, data properties, and output shapes to identify the exact optimal pattern in under 30 seconds.
- Input size N instantly eliminates infeasible time complexities: N <= 20 allows O(2^N) backtracking/bitmasking; N <= 10^5 requires O(N log N) or O(N).
- Sorted arrays or monotonic predicates point directly to Two Pointers or Binary Search on Answer Spaces.
SYSTEM DESIGN FUNDAMENTALS8 min read · 2026-11-20
How to run clean 3-minute capacity estimates in L5 Senior and L6 Staff system design rounds—and use those numbers to justify sharding, caching, and queues.
- 1 day = 86,400 seconds (~10^5 seconds). 1 million requests/day is ~12 RPS average; 100 million requests/day is ~1,150 RPS average.
- Never stop at arithmetic: every estimate must drive an architectural decision (e.g., 45,000 write QPS exceeds a single primary Postgres node -> partition by shard key).
ARCHITECTURE CASE STUDIES10 min read · 2026-11-20
A Staff-level walkthrough of payment processing interviews at Stripe, Amazon, Coinbase, and Uber: preventing double charges, state machines, and PSP webhook reconciliation.
- Network timeouts are ambiguous: the client cannot tell whether the server crashed before charging or after charging.
- Client-generated UUIDv4/v7 idempotency keys enforced by a database UNIQUE constraint turn at-least-once network retries into exactly-once business execution.
GENAI & LLM SYSTEMS11 min read · 2026-11-20
How to pass the modern GenAI architecture round at OpenAI, Anthropic, Meta, and Google: TTFT vs. TPOT budgets, continuous batching, hybrid vector search, and guardrails.
- Separate LLM latency into Time-To-First-Token (TTFT < 250ms, compute-bound prefill) and Time-Per-Output-Token (TPOT < 30ms/token, memory-bandwidth-bound decode) streamed over SSE.
- Prefix-aware L7 routing directs multi-turn chat turns to the GPU replica holding warm PagedAttention KV-cache blocks, slashing prefill FLOPs by up to 80%.
COMPANY LOOP PLAYBOOKS8 min read · 2026-11-20
Inside Meta's E4/E5 technical loop: executing two bug-free optimal problems in 40 minutes without running code, plus Product and Infrastructure system design expectations.
- Meta coding interviews give 45 minutes total (~38–40 minutes of coding time) to solve TWO medium/medium-hard problems completely—including manual line-by-line dry runs.
- Code execution is disabled in CoderPad during Meta coding screens: you must trace boundary inputs manually before declaring victory.
COMPANY LOOP PLAYBOOKS8 min read · 2026-11-20
How Amazon interviewers and Bar Raisers grade coding maintainability, distributed system resilience, SQS/Kafka backpressure, and Leadership Principle depth.
- Amazon Bar Raisers probe what breaks at 3 AM: poison-pill messages, downstream thundering herds, AZ network partitions, and retry storms.
- Pair asynchronous queues with Dead-Letter Queues (DLQs), jittered exponential backoff, circuit breakers, and token-bucket rate limiters.
COMPANY LOOP PLAYBOOKS8 min read · 2026-11-20
How to handle disguised algorithmic prompts in Google coding loops—where standard patterns hide behind novel domain stories—plus L5+ infrastructure design.
- Google rarely pastes verbatim LeetCode titles; prompts wrap Graphs, Dijkstra, Union-Find, Tries, or Binary Search on Answer Spaces inside custom engineering scenarios.
- Expect progressive 2- or 3-part follow-ups (e.g., single query -> 10^5 streaming queries -> memory-constrained distributed version).
ARCHITECTURE CASE STUDIES9 min read · 2026-11-20
How to architect a 500,000-write/sec real-time location ingestion pipeline, spatial driver lookup index, and atomic ride-matching state machine.
- Read-heavy business directories (Yelp) suit static Quadtrees or read-replicated Geohash indexes; Uber's 4-second GPS pings from millions of active drivers create a massive write-heavy spatial workload.
- Separate ephemeral driver location state (in-memory sharded Redis Geohash / H3 hexagonal cells) from durable trip & billing state (ACID sharded SQL).