Content Moderation Pipeline

Interview-Ready Approach

1) Clarify Scope and SLOs

• •Problem statement: Design a content moderation system that screens text, images, and video using AI + human review for a social platform.
• •Design for a peak load target around 579 RPS (including burst headroom).
• •Content items/day: ~10,000,000
• •Auto-moderation latency: < 30 seconds
• •Human review queue time: < 30 minutes
• •False positive rate: < 1%
• •False negative rate: < 0.1%

2) Capacity Planning Method

• •Convert traffic and growth constraints into request rate, storage growth, and concurrency budgets.
• •Keep at least 2-3x safety margin per tier (ingress, compute, storage, async workers).
• •Reserve explicit latency budgets per hop so p95 can be defended in review.

3) Architecture Decisions

• •Message Queues: Move non-blocking and retry-heavy work to async consumers with explicit retry and DLQ policies.
• •Databases: Define a clear system-of-record and design read/write paths separately before adding optimizations.
• •API Design: Standardize API boundaries, idempotency keys, pagination, and error contracts first.
• •Storage: Use object storage for large blobs and keep metadata/authorization separate in the API tier.
• •Monitoring: Instrument golden signals (latency, traffic, errors, saturation) per tier and per tenant/domain.

4) Reliability and Failure Strategy

• •Guarantee idempotent consumers and trace every message with correlation IDs.
• •Use strong write constraints (transactions or conditional writes) and explicit backup/restore strategy.
• •Apply strict input validation and backward-compatible versioning.
• •Enforce lifecycle policies, retention tiers, and checksum validation.
• •Alert on user-impact SLOs, not only infrastructure metrics.

5) Validation Plan

• •Run one peak-load test, one dependency-degradation test, and one failover test.
• •Verify idempotency for all retried writes and async consumers.
• •Track user-facing SLOs first: p95 latency, error rate, and successful throughput.

6) Trade-offs to Call Out in Interviews

• •Message Queues: Async pipelines absorb spikes well, but increase eventual-consistency complexity.
• •Databases: SQL gives stronger transactional guarantees; NoSQL often gives better write scaling and flexibility.
• •API Design: Rich APIs improve developer speed but can create long-term compatibility burden.
• •Storage: Object storage is cheap and durable, but random low-latency reads are weaker than databases/caches.
• •Monitoring: Deep observability speeds incident response but raises ingestion and tooling costs.

Practical Notes

• •Design as a pipeline: upload → text classifier → image classifier → decision engine → publish or queue for human review.
• •Use a message queue between stages so each classifier can scale independently.
• •Human review: distribute items round-robin with a maximum queue depth per reviewer. Track inter-rater reliability.

Why This Solution Works

Request path: The solution keeps ingress, service logic, and stateful dependencies separated so each layer can scale independently.

Reference flow: Web Clients -> Load Balancer -> API Gateway -> API Service -> Primary SQL DB -> Message Queue -> Background Workers -> Object Storage

Design strengths

• •Async queue/event bus isolates bursty workloads and supports retries without blocking synchronous requests.
• •Monitoring and logs are wired in from day one for rapid incident triage.

Interview defense

• •This design makes bottlenecks explicit (ingress, core compute, persistence, async workers).
• •It supports progressive scaling without re-architecting the core request path.
• •It keeps correctness-sensitive state changes in durable systems while offloading background work asynchronously.