Databricks AI Summit 2023 Grammarly Session
Grammarly faced scalability issues with its in-house data platform, Gnar, and embarked on a redesign called the Event Store. They adopted technologies like Spark Structured Streaming, Delta Lake, and Databricks Unity Catalog. The migration process involved merging event schemas, rebuilding ingestion and storage layers, and backfilling data from 2015 to the present. Lessons learned included the importance of enforcing schemas early on and the challenges of backfilling. The redesign improved query performance, enabled feature creation, and reduced data availability time. Overall, Grammarly now has a more scalable and efficient data platform.