How a one-person data team at Keep built real-time banking infrastructure on dbt

Keep’s head of data, Scott Ziolko, described how a one-person data team replaced a manual, spreadsheet-based credit underwriting process that took up to a business day with an automated system delivering post-ingestion decisions in under a minute. Serving Canadian small businesses, Keep chose managed tools rather than building infrastructure internally, using Artie for sub-second change data capture from Postgres and MongoDB into Snowflake, Snowflake dynamic tables, streams, tasks, and stored procedures for event-driven processing, and dbt for model governance, versioning, deduplication, snapshots, and non-real-time transformations. Banking data collected through Flinks is replicated, flattened, and transformed into cash-flow and risk features, while Streamlit and Snowflake provide a governed operational interface for underwriting teams. Ziolko noted that real-time components sacrifice some of dbt’s usual observability, so Keep uses custom freshness tests and retains every underwriting calculation with its producing Git hash for traceability and portfolio-wide model reprocessing. Although downstream processing is reliably under a minute, external bank connections can take longer, illustrating that end-to-end latency also depends on vendor performance and cost decisions.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.