How ERGO Hestia reduced time-to-market with Databricks Lakebase and Model Serving

Summary

ERGO Hestia redesigned its real-time pricing platform to reduce deployment friction across more than 100 models and 1,000 variables while preparing B2C capabilities. Previously, processed data moved from Databricks through extraction jobs, external Azure PostgreSQL, and a custom caching adapter, creating governance overhead, deployment coordination, and latency spikes during large refreshes. The new architecture uses Lakebase Sync Tables as an online serving layer and Databricks Model Serving Endpoints, keeping data, request logic, and model serving within the lakehouse; Unity Catalog supplies lineage, version tracking, access controls, and audit trails. An incremental migration started with a low-criticality endpoint, measuring 20ms latency and less than 5% CPU utilization at 40 requests per second, before expanding toward larger workloads and the planned decommissioning of PostgreSQL.

Context

ERGO Hestia needed to support real-time B2C pricing and faster model updates as its pricing platform grew beyond 100 models and 1,000 variables. Its previous architecture depended on external Azure PostgreSQL, extraction jobs, a caching layer, and an adapter, which increased operational overhead, complicated governance, constrained deployment timing, and caused 10x to 20x latency spikes during large data refreshes.

Approach / What changed

The team moved data serving into the Databricks lakehouse using Lakebase Sync Tables and exposed pricing data and models through Databricks Model Serving Endpoints. Existing Databricks ETL pipelines and MLflow-registered models were retained, while Unity Catalog unified governance, lineage, versioning, and access controls. Migration proceeded incrementally, beginning with a low-criticality endpoint and validating performance before expanding to larger workloads.

Takeaways

• Lakebase Sync Tables continuously synchronize processed Delta data with the serving layer, removing manual extraction jobs and the external PostgreSQL dependency from the serving path.
• The proof of concept recorded 20ms latency and less than 5% CPU utilization under a load of 40 requests per second.
• Unity Catalog connects data and model governance, providing lineage, version tracking, access controls, and audit trails for pricing decisions.

Related reading