Fantasy Football AI
Gameweek intelligence for Fantasy Premier League.
Transfers, captaincy, chips, live points and mini-leagues. One clear plan per deadline.
Open the app →
What you get
Enter your public FPL Team ID. No login is needed, and the app never changes
your team.
| Feature | What it does |
|---|
| ✅ Gameweek to-do | Home opens with this week's list: the transfer (or roll), the captain, the chip call and the player to check, with a confidence level. The plan is ready as soon as the previous deadline passes. |
| 📡 Live | Live points against the model's expectation. Tap any player for their points breakdown. |
| 🔁 Transfers | Every option compared on one scale, a two-week plan (move now or wait?) and Try your own swap: is your idea better than the plan? |
| 👑 Captain | A realistic points range for each candidate, and a pick that weighs safe against differential for your mini-league position. |
| 🃏 Chips | Play now or save, a season chip plan around blank and double Gameweeks, and the best Free Hit and Wildcard squads right now. |
| 🏆 Mini-leagues | The race, a head-to-head against any rival, and the players your league relies on. |
| 💷 Prices | Who is likely to rise or drop tonight, and price-smart swaps that do so without losing points. |
| 📊 Explore & History | Every player's forecast, xG and xA per 90, fixture runs with clean-sheet odds, and a season review against the model. |
| 📱 Install | Add it to your Home Screen and it opens like an app. |
Rival team names and the league name in these screenshots are anonymised.
See every feature with screenshots →
How it works
flowchart LR
FPL[Official FPL API] --> API[FastAPI backend]
CSV[Published forecast] --> API
API --> ENG[Decision engine]
ENG --> API
API --> WEB[Next.js app]
GHA[GitHub Actions<br/>every 30 minutes] --> CSV
- Forecast: a gradient-boosting model of each player's points. In the
inspected 2026/27 GW1–5 evaluation its mean absolute error was 1.26, against
1.35 for a last-five average. It is re-checked after every Gameweek; future
improvements require evaluation on previously reserved rounds.
- Decisions: one optimiser picks the squad, XI and captain over five
weighted Gameweeks. Holding, free transfers and hits are scored on one
scale.
- Honest labels: every number's source (FPL data, Model or Derived) is
labelled, once per card, and uncertainty is shown, never hidden.
More detail: architecture ·
model · API
Run it locally
git clone https://github.com/GAWI01/fantasy-football-ai.git
cd fantasy-football-ai
docker compose up --build # app on :3000, API on :8000
python -m venv .venv && source .venv/bin/activate
python -m pip install -r requirements.txt -r backend/requirements.txt
python -m uvicorn backend.main:app --reload --port 8000
In a second terminal:
cd frontend
corepack enable && pnpm install --frozen-lockfile
cp ../.env.example .env.local
pnpm dev
| Variable | Purpose | Default |
|---|
NEXT_PUBLIC_API_URL | Backend origin, compiled into the app | http://127.0.0.1:8000 |
FPL_AI_CORS_ORIGINS | App origins the API accepts | http://localhost:3000,http://127.0.0.1:3000 |
python -m pytest # backend, decision engine, model pipeline
python -m benchmark.run # decision benchmark on recorded real teams
cd frontend && pnpm test && pnpm lint && pnpm exec tsc --noEmit
Docs
Data and credits
- Historical data: the model is trained on past FPL seasons from
vaastav/Fantasy-Premier-League
by Vaastav Anand, used with his permission and this attribution. Thank you!
- Live data: the public Fantasy Premier League API.
Free, independent and non-commercial. Not affiliated with or endorsed by
the Premier League, Fantasy Premier League or any football club. FPL data,
club names, colours and marks belong to their owners; kits are simplified
drawings made for this site. Code under the MIT licence; data is not covered,
see DATA_NOTICE.md. Made by GAWI01.
