Data Architecture Foundations

You can run a database, build a pipeline and design a star. This course is about the decisions that come before all three: which store, which model, which integration pattern, what the retention and security design has to be, and how to write that down so another engineer can challenge it. You work on a three-store estate on your own lab machine — **Ultishop**, an order system of record; **Riverbank Media**, a streaming platform with a landing, clean, warehouse and quarantine pipeline; and **Meridian**, a finished retail star you cannot even read as yourself. Nothing is taken on trust. The inventory is collected from the stores' own catalogs. The storage-model argument is settled by modelling the same catalogue normalised and as documents and measuring what each costs, including the measurement that decides it, which is a write rather than a read. The warehouse-style argument is settled by building a small Data Vault and loading two billing snapshots through it. Then the platform gets built. Change data capture from the order system to a second PostgreSQL cluster, with a replication role that can read three tables and a password that never enters the catalog — and a seeded fault that stops the apply worker dead, so you diagnose a stalled stream from the catalog and the log rather than from a diagram. A partitioned open-format lake, read back through a foreign table. A catalog and column-level lineage that is validated against the database it describes. A certificate that a client can actually verify, field encryption whose key never enters the database, a masked view tested from both directions, and retention applied by dropping partitions — with a certificate change that stops the server from starting, because that is the one you want to have rehearsed. The last lesson is the work you will be given most often: somebody else's package, with a proof of concept that loads and returns correct numbers, and five claims in its document that are not true. You review it against a checklist and record every finding as a query anybody can re-run. The final project is an engagement for a four-site veterinary practice with clinical, personal, activity and financial data, delivered as an architecture package plus a proof of concept whose checks pass.