Scaled to 1M+ users

I co-founded BoringRats and owned the backend and data infrastructure from the first monolith to a horizontally scaled system serving 1M+ active users.

• Event-driven architecture on Kafka with idempotent consumers, so messages were neither lost nor processed twice
• Real-time analytics pipeline (Airflow and Spark) feeding product decisions with sub-minute freshness
• Read-heavy API layer tuned with Redis caching and connection pooling, holding p99 latency under 250ms
• Zero-downtime deploys with canary releases, so shipping never interrupted users

The hard part was the growth curve. What worked at 10K users broke at 100K, and I spent months on capacity planning, backpressure, and database read-path optimization. Designing for ten times the load you have today stopped feeling like overkill.