Stepan Pushkarev Interview
summary
Hydrosphere.io focuses on shipping ML pipelines into production and reports concrete engineering choices: using a functional paradigm so models are delivered as pure function deployment units to reduce side effects, building microservices that minimize data movement with flatbuffers for IPC, and exploring Spark Compute as a Service and cost-effective GPU use for inference. The main surprise is that data transfer (large images, binaries) becomes the scaling bottleneck more often than raw compute.
AI-written summary · about
