Analyses, rendered from
Markdown, SQL and AI.

Real, interactive dashboards — charts from SQL, the analysis written by an LLM, all authored in plain .md files and exported to a static site. Explore one, then build your own.

tour-de-france.dashdown.ai

The battle for yellow refreshed daily · 08:00 UTC

gc gap — raced predicted (ML) route elevation

// live + ml · tdf 2026

Tour de France 2026: The Machine Learning Grand Tour

Real standings and stage results so far — and ML predictions for every remaining stage, the final podium and all four jerseys, re-fetched and re-trained daily at 08:00 UTC while the race runs. With an honest model scorecard, cross-validated on held-out years.

The Battle for Yellow Stage-by-Stage Predictions Final Podium Forecast Jersey Projections Model Scorecard

View the analysis →

nyc-taxi-rides.dashdown.ai

Monthly trips 2020 → now

yellow + green cab Uber + Lyft

// scale · parquet + duckdb

1.6 billion NYC rides, straight from raw Parquet

Every yellow cab, green cab, Uber and Lyft trip since 2020 — 1.6B rows of NYC TLC Parquet, queried directly with DuckDB and rebuilt nightly by GitHub Actions. No ETL, no warehouse, no server: one duck, zero servers.

Taxis vs Ride-Hail Financial Metrics Fares & Tips Newest Month Monthly Aggregates

View the analysis →

ai-boom.dashdown.ai

Monthly AI-package installs 2019 → 2026

frameworks provider SDKs infrastructure

// data story · PyPI

Watching the AI boom in 150 billion PyPI downloads

42 AI packages tracked from 2019 to 2026, reading pip installs like a seismograph — the framework war, the compute tell, hype vs. real usage, and LLM-written takeaways. Every number and paragraph baked from Markdown + SQL.

The Seismograph The Framework War The Compute Tell Hype vs. Usage AI's Read on Data

View the analysis →

boring-semantic-layer.dashdown.ai

Revenue over time May 2005 → Feb 2006

revenue

$67.4k

rentals

16,044

customers

599

metric={total_revenue} by={month} · no SQL on the page

// semantic layer · sakila

16,000 video rentals, zero SQL on the page

The classic Sakila rental database served through the Boring Semantic Layer: metrics and dimensions defined once in a YAML model, every widget just metric={…} by={…} references. Pick a filter and charts, counters and the AI commentary all re-query through the same definitions.

Revenue over Time What Sells, and Where The Month in Numbers Films & Customers How It Works

View the analysis →

steam-graveyard.dashdown.ai

Every game, by lifetime reviews log-width buckets

ghosts — under 50 reviews found an audience

// self-narrating · steam

(Almost) every game on Steam. Most of them are ghosts.

A self-narrating dashboard on the brutal economics of the world's biggest PC games store. Every ✦ paragraph was written by an LLM reading the query results at build time, then frozen into static files — and a GitHub Action re-crawls SteamSpy and Steam's live review endpoint weekly.

The Flood The Power Law The Graveyard Where the Odds Are Least Bad The Price Paradox

View the analysis →