Databricks AI Summit 2023 Grammarly Session

• Scaling Event Processing and Analytics in the Data Lake

Grammarly's in-house data platform, Gnar, stopped scaling, so they redesigned it as the Event Store, on Spark Structured Streaming, Delta Lake, and Databricks Unity Catalog. The migration meant merging event schemas, rebuilding the ingestion and storage layers, and backfilling data from 2015 to now. Their lessons were to enforce schemas early, and that backfilling is hard. Queries got faster, building features got easier, and data shows up sooner. Grammarly now has a data platform that scales.