Blog
Para quem já sabe o básico e quer ir fundo. Aqui o assunto é como os modelos funcionam em produção: memória, roteamento, ferramentas, agentes. O lado técnico que pouca gente explica direito.
FRAME: Functional requirements, Non-functional (scale, latency, consistency), API, Model (data), Estimate. 5 passos que estruturam qualquer interview.
Latência (L1 0.5ns, disk 10ms, cross-region 150ms). Throughput (QPS por tipo). Storage (MB/record). Cost napkin. Powers of 10 + memorizados.
Design de bit.ly: hash vs counter, base62, DB schema, cache layer, analytics. Trade-offs consistency, hot shortcuts.
Fan-out on write vs read vs hybrid. Celebrity problem. Pull model. Feed ranking. Timeline storage. Cache warming.
Token bucket, leaky bucket, sliding window (counter + log). Redis-based (lua script atomic). Consistent vs approximate. Cross-region.
WebSocket connections, long polling fallback. Delivery semantics (at-least-once + idempotency). Read receipts. Group chat fan-out. Encryption at-rest + E2E.
Push (APNs/FCM), email (SES/Sendgrid), SMS. Templates, user preferences, quiet hours. Priority queues, rate limiting por user. Delivery tracking.
Hash ring (consistent hashing), replication, failover. Cache patterns (aside, through, back). Eviction policies. Hot keys, thundering herd.
Inverted index, tokenization, ranking (BM25 + learning-to-rank), query suggest, autocomplete. Elasticsearch arch. Vector search hybrid.
Simular interview 45min: dropbox-like OU booking-like OU youtube-like. Usar framework. Entregáveis: diagrama + back-of-envelope + trade-offs escritos.