// Link · 3 February 2026

How ShareChat Scaled their ML Feature Store 1000X without Scaling the Database

ShareChat engineers rebuilt a failing ML feature store into a system capable of serving billions of features per second—without scaling the database. By redesigning data models, optimizing tiling, improving cache locality, and tuning gRPC and GC behavior, they turned ScyllaDB into a low-latency back...

Hacker Noon

@hacker-noon · hackernoon.com

hackernoon.com

Hacker Noon@hacker-noon