Quantified Self 2.0: Build a Unified Health Data Warehouse with DuckDB and dbt

Ever tried to compare your Oura sleep score with your Garmin body battery, only to realize you’re comparing apples to... well, very differently formatted oranges? If you're a data nerd like me, you probably track everything. But the "Quantified Self" dream quickly turns into a nightmare when you're juggling JSON exports from Oura, messy CSVs from Fitbit, and timezone-conflicted data from Garmin. In this tutorial, we are going to solve the fragmented data problem by building a local Modern Data Stack (MDS) . We’ll use DuckDB as our powerhouse engine and dbt (data build tool) to transform raw, messy health metrics into a standardized Common Data Model (CDM) . By the end of this, you'll have a production-grade Data Engineering pipeline running right on your laptop, ready for advanced AI analysis or visualization in Apache Superset . The Architecture The goal is to move from "Raw Silos" to a "Unified Analytics Layer." We use Python to fetch/land the data, DuckDB as the storage and compute