Inspiration
A “small” schema change can silently break dashboards, models, and ML pipelines across an organization. We built Cascade to turn that hidden blast radius into a coordinated, auditable migration.
What it does
Cascade detects breaking schema changes, traces affected assets through DataHub lineage, rewrites downstream SQL, validates the result, opens remediation pull requests, and writes the decision trail back to DataHub.
How we built it
We combined a Python CLI and FastAPI backend with DataHub’s metadata graph, GitHub Actions, and a React/Vite interface. LLM-powered rewrites are schema-gated, with deterministic fallbacks for reliable offline execution.
Challenges we ran into
The hardest problems we faced were translating code diffs into schema changes, safely rewriting downstream SQL, handling incomplete lineage, and making every GitHub and DataHub action idempotent and auditable.
Accomplishments that we're proud of
We built a complete impact-to-remediation loop: blast-radius analysis, severity scoring, SQL patches, remediation PRs, policy checks, DataHub write-backs, and an interactive visual demo.
What we learned
Metadata becomes far more valuable when agents can both reason over it and improve it. Reliable automation also requires strong validation, deterministic fallbacks, and a clear audit trail, not just generated code.
What's next for Cascade
We plan to publish Cascade on PyPI as an easy-to-install package, making it simple for teams to integrate into their own repositories. We also want to support multiple databases, including Postgres, Snowflake, and BigQuery.
Built With
- datahub
- fastapi
- github
- githubactions
- postgresql
- python
- react
- typescript
- vercel
- vite
