System Design
Interactive architecture breakdowns of systems used by billions. Explore animated data flows, component trade-offs, and the exact tech stack behind each design.
End-to-end encrypted messaging at 2 billion users. From client to XMPP gateway to Cassandra-backed message store — see every hop in the message journey.
Global streaming at 270M subscribers. Adaptive bitrate delivery, multi-region CDN, recommendation engine, and the Open Connect backbone powering every play.
500 hours of video uploaded every minute. Explore the upload pipeline, transcoding farm, global CDN, and recommendation system that drives 70% of watch time.
2 billion monthly users sharing 100M+ photos daily. See how the feed ranking, media pipeline, follower graph, and Stories infrastructure all fit together.
Multi-agent AI orchestration — how LLMs plan, use tools, call APIs, manage memory, and chain reasoning steps. The architecture behind autonomous AI systems.
Simulate Round Robin, Weighted RR, Least Connections, and IP Hash algorithms. Watch requests fly across servers in real time with per-server load meters.
Phase 1Tune cache capacity, eviction policy (LRU / LFU / FIFO), and access pattern. Watch hit rates and average latency update live as your cache warms up.
Phase 1Step through 1NF → 2NF → 3NF normalization with real table examples. See redundancy shrink and understand update/insert/delete anomalies visually.
Phase 1Compare full table scan vs B-Tree index traversal on tables from 1K to 1M rows. See the O(n) vs O(log n) difference play out visually with rows-scanned counters.
Phase 1Spin up 1–4 replicas and toggle Sync / Async / Semi-Sync modes. Watch replication lag appear in async, fail a replica to test HA, and route reads to replicas.
Phase 1Distribute records across 2–6 shards with Hash, Range, or Directory strategies. Type any key to see exactly which shard it lands on — and watch hot shards turn red.
Place servers on a virtual ring with N virtual nodes each. Add/remove servers and watch only the minimal keys remap — no global reshuffle needed.
Phase 2Inject a network partition and watch CP systems reject reads while AP systems serve stale data. See why you can only pick 2 of 3 guarantees.
Phase 2Kafka vs RabbitMQ — tune producers, partitions, and consumer groups. Simulate lag building up and watch consumers fall behind in real time.
Phase 2Publish events to a bus and watch multiple subscribers react in parallel. Toggle comparison mode to see request-response vs event-driven latency side by side.
Phase 2Watch services compete for a Redis lock with TTL. Crash the holder mid-operation and see the lock expire and transfer. Toggle to Redlock for 5-node quorum mode.
Phase 2Run a saga across Order → Payment → Inventory → Shipping. Inject a failure at any step and watch compensating transactions roll back the whole saga automatically.
Token Bucket, Leaky Bucket, Fixed Window, Sliding Window — watch requests get accepted or throttled with a live 429 rejection counter and fill-level gauge.
Phase 3Set failure probability and compare No Retry vs Fixed vs Exponential vs Jitter. See the thundering herd spike collapse into a smooth load curve when jitter is added.
Phase 3Tune failure threshold and recovery timeout. Watch the circuit flip Closed → Open → Half-Open with a live state machine and failure-rate gauge.
Phase 3Send 3–5 duplicate payment requests with and without an idempotency key. See duplicates pile up on one side and get deduplicated on the other — side by side.
Toggle Auth, Rate Limiting, Caching, and Transform features. Watch each feature light up as requests flow through the gateway pipeline — inject failures to see 401, 429, and 503 responses.
Phase 4Scale instances up and down, kill one mid-flight, and watch the registry detect the health check failure and route around it. Toggle client-side vs server-side discovery.
Phase 4Step through JWT auth, OAuth 2.0 Authorization Code, and Client Credentials flows. See the JWT decoded live, simulate token expiry, and watch the refresh token flow trigger.
Phase 4Inject error spikes, latency spikes, or memory leaks. Watch alerts fire on the metrics chart, ERROR lines stream in the log feed, and a slow span appear in the distributed trace waterfall.