At Patreon, I re-architected the real-time ingestion layer in C++20, replacing a Python service and handling a Kafka feed of 250K+ events per second. The change reduced p99 processing latency from 40ms to 6ms.
I built multithreaded consumers with lock-free queues and tuned memory and CPU-cache behavior, cutting per-event CPU cost by half under sustained peak load. I also hardened the pipeline against backpressure with bounded queues and graceful shed policies.
At Spotify, I designed C++17 stream processors for real-time playback telemetry and built concurrency primitives that raised throughput on fixed hardware by roughly 40%. I also developed protocol adapters and schema validation to support varied upstream sources and evolving event structures.
Earlier, at Tumblr, I developed and maintained high-volume data ingestion services for real-time activity streams, optimizing connectors and processing code to reduce batch latency and CPU cost on high-traffic feeds.
