BharatMLStack

Why we included this project

BharatMLStack is the ML infrastructure platform Meesho built to serve features, run model inference, and handle embedding search at high request rates, shipping as a set of Kubernetes-native components rather than a single monolith. The online feature store returns features in under 10ms, Inferflow orchestrates DAG-based inference pipelines, and Skye provides vector similarity search with pluggable backends, all coordinated by a control plane and surfaced through a web console for feature registry and approval workflows. Since it is cloud-agnostic and comes with Go and Python SDKs plus a Docker Compose quick start, you can try it locally before committing to a full deployment. For teams that need low-latency feature serving and inference at scale without stitching together half a dozen services, this is a practical reference to start from.