This is a submission for Weekend Challenge: Dog Days Edition
What I Built
Spurious Bark 🐕 is a Streamlit app that answers a question nobody asked: does German dog tax revenue secretly control the rest of society?
You pick a second time series from a dropdown — marriages, beer production, whatever I could scrounge from official German statistics — and the app plots it next to dog tax revenue, calculates the real statistical correlation between the two using Snowflake, and then asks Snowflake Cortex to generate a completely fake, dead-serious "scientific explanation" for why they're connected. Think tylervigen.com/spurious-correlations, but with more tail-wagging.
Turns out dog tax revenue and beer production correlate at -0.836. Coincidence? Yes. Obviously yes. But Cortex will still try to convince you otherwise.
Demo
The app compares dog tax revenue against:
- 💍 Marriages
- 🍺 Beer production (in hectoliters, because Germany)
Both time series come from the German Federal Statistical Office (Destatis), loaded into Snowflake tables and joined live via SQL.
Code
How I Built It
The stack is deliberately small: a local Streamlit app talking directly to Snowflake, no Streamlit-in-Snowflake, no extra services.
-
Data: dog tax revenue plus two comparison series (marriages, beer production), all pulled from Destatis' GENESIS-Online database and loaded into two Snowflake tables (
dog_tax_revenue,comparison_series) -
Correlation: computed live in Snowflake with a plain
CORR()aggregate over a joined query — no pandas math, the database does the work -
The punchline:
SNOWFLAKE.CORTEX.COMPLETE()takes the correlation coefficient and the series name, and generates a short, silly, deliberately-fake explanation in the spirit of "spurious correlations" — then explicitly tells the reader it's nonsense, because correlation ≠ causation -
Making it demo-proof: two things I added once I hit real-world snags —
- A demo mode that kicks in automatically if no Snowflake credentials are configured, using generated sample data, so the app is inspectable without any setup
- A graceful fallback for the Cortex call: my trial account turned out to have Cortex functions locked (a known trial-account limitation), so instead of crashing mid-demo, the app catches that and falls back to a canned example explanation with a clear "Cortex was unreachable" note — because nothing kills a hackathon demo like an uncaught exception in front of the judges
- The look: bright, rounded, dog-themed UI (yes, there are paw print dividers) because a project about dogs should not look like a quarterly earnings report
The comparison_series table, live in Snowflake — 17 rows of real German government data, no mocks.
Prize Categories
Submitting for Best use of Snowflake 🐾 — the entire correlation calculation runs as SQL inside Snowflake (CORR()), and the comedic core of the app is powered by SNOWFLAKE.CORTEX.COMPLETE() generating the fake explanations live.
Top comments (1)
Wow, I am enlightened. I didn't knew German dog gave taxes. Cool concept. 😁