Skip to content

About

F1 race winner predictions using a 12-feature weighted model with softmax. Historical data from 2000–2025 via FastF1. Built with Astro SSR, Hono, Drizzle ORM, and Neon PostgreSQL on Cloudflare.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

3 stars

Watchers

0 watching

Forks

F1 Intelligence Platform

Every race, explained. The next one, predicted.

Race breakdowns, championship odds, history back to 1950 and a fitted model that gives every driver a chance to win, before and after qualifying.

Open the site · How the model works · Docs · Contributing · Report a bug

test License: GPL v3 Runs on free tiers

The home page: season progress, the next race with its countdown and pre-qualifying favourites, the last race's podium and model check, and the championship top 5

Features

  • Home: where the season stands, the next race with a countdown and the model's top 3, the last race's podium and recap, and the title fight
  • Races: results, qualifying, lap charts, practice long runs, tyre strategy, race control, overtakes and team radio
  • Race story: gap to the leader, position changes, battles and weather, lap by lap from 2018 on
  • Telemetry: two drivers' fastest qualifying laps side by side, with speed, delta and mini-sector dominance
  • Predictions: a win chance for every driver at every GP and sprint, why the model picked its winner, and a season scorecard
  • Championship: driver and constructor standings, title odds after every round and a what-if calculator
  • Circuits: track maps, overtaking and safety car rates, past winners and a live weekend forecast
  • Compare: any two drivers or teams, by season or across whole careers
  • History: all-time records, milestones, on this day and constructor lineages back to 1950

Screenshots

Race story Qualifying telemetry
Gap-to-leader and position charts for the top 10 finishers of the United States Grand Prix Verstappen and Hamilton's fastest Mexico City qualifying laps compared by speed and mini-sector
Prediction scorecard Circuit
The 2026 scorecard: hit rates, best call, worst miss and every race's pick against its winner The Las Vegas Strip Circuit with its track map, overtake rate, safety car stats and weekend forecast
History Title odds
All-time records with the current drivers chasing each one The title fight: each driver's title odds, likely final points and odds by round

More pages in preview/.

How the predictions work

Once qualifying is done, the model scores every driver on seven things it knows before the lights go out: grid slot, how hard the track is to pass on, recent form, the team's race and qualifying pace, this weekend's qualifying gap and the gap to the team-mate. The weights aren't hand-picked: they are fitted to every Grand Prix since 2010, and a softmax turns the scores into a win chance for each driver.

Replayed race by race on 190 Grands Prix (2018–2026), using only what was known before each one, it picks the winner 60% of the time and has them in its top 3 90% of the time. Always backing the pole-sitter gets 55%.

  • Sprints use the same weights on the sprint grid
  • Before qualifying, a forecast from form and team pace stands all week, and the page shows how qualifying moved each driver
  • Title odds come from 10,000 simulated rest-of-seasons after every round

Everything else (features, backtests, the models it replaced): docs/prediction-model.md.

Data

Live sessions come from OpenF1 and Jolpica, both free with no account. All-time history comes from F1DB (CC-BY-4.0), weekend forecasts from Open-Meteo. Race results go back to 1980, lap-level data to 2018. What comes from where, and how far back: docs/data-sources.md.

Tech stack

Frontend Astro (server-rendered) and Tailwind on Cloudflare Pages; plain SVG charts
API Hono and Drizzle ORM on Cloudflare Workers, cached at the edge
Data engine Python 3.12 ETL worker on Render: ingest, features, predictions, title odds
Database Neon PostgreSQL
Hosting cost $0: every layer runs on a free tier and is built to stay inside it

Running locally

You need Bun, Python 3.12 and Docker. Local dev runs against a local Postgres snapshot, never the production database:

git clone https://github.com/Glory42/F1-Intelligence-Platform.git
cd F1-Intelligence-Platform
bun run install:all
bun run dev   # API on :8787, site on :4321

Filling the local database, the data engine, tests and checks: CONTRIBUTING.md.

Project layout

apps/
├── web/        Astro site
├── api/        Hono API, Drizzle schema and migrations
├── e2e/        Playwright smoke tests
└── og/         Share-image renderer
data-engine/    Python ETL worker, backfills, backtest and model fit
docs/           Architecture, pipeline, schema, model, API, deployment

Documentation

Read this For
CONTRIBUTING.md Setup, tests, conventions, the rules every change follows
docs/architecture.md How the site, API, data engine and database fit together
docs/data-sources.md Where the data comes from and what each era has
docs/data-pipeline.md Every ETL job, the race-weekend schedule, backfills
docs/prediction-model.md Features, fitted weights, backtests, title odds
docs/api-reference.md Every API route and its response
docs/database-schema.md Tables, keys and conventions
docs/frontend.md Pages, components and the server-only data rule
docs/testing.md The five test suites and how CI runs them
docs/deployment.md Cloudflare, Render, environment variables, free-tier limits
CODEMAP.md File-by-file map of the repo
CONTEXT.md and docs/adr Domain language and architecture decisions

App-level guides: apps/api · apps/web · data-engine.

Contributing

Issues and pull requests are welcome: start with CONTRIBUTING.md. Wrong data on a page? Use the data issue template. By taking part you agree to the Code of Conduct. Please report security issues privately as described in SECURITY.md.

License

GPL-3.0. Historical data from F1DB under CC-BY-4.0. This is an unofficial fan project and isn't associated with the Formula 1 companies. F1, FORMULA ONE and related marks are trademarks of Formula One Licensing B.V.

About

F1 race winner predictions using a 12-feature weighted model with softmax. Historical data from 2000–2025 via FastF1. Built with Astro SSR, Hono, Drizzle ORM, and Neon PostgreSQL on Cloudflare.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

3 stars

Watchers

0 watching

Forks

Used by

Contributors

Languages