I wanted a reference platform I could hand to a client's architect: real-time telemetry at a volume that makes design mistakes visible, built the way I would build it on an engagement.
A production-grade medallion lakehouse on Databricks processing telemetry from 2.4 million IoT smart meters, with Terraform modules, dbt models, and an architecture-as-code pipeline that keeps the documentation in step with what is deployed.
Why build it
Every data engineer can say “medallion architecture”. Far fewer can show you one at a volume where a wrong partitioning or merge strategy shows up in runtime and cost. This exists so a client’s architect can read the code instead of taking my word for it.
What is in it
Bronze, silver and gold layers on Delta, dbt models on top, and Terraform modules for the infrastructure. The architecture diagram is generated from the deployed state, so the documentation cannot quietly drift from reality.
Honest limits
It is a reference implementation with synthetic telemetry, not a client system with a support rota behind it. The design is the deliverable, and it is public: the code is on GitHub.