Built for serious EPL watchers

Stop guessing.
Start reading the game properly.

Every weekend millions of EPL decisions are made on gut feel, pundit takes, and Twitter threads. Castle Footy AI exists to give you something better — structured analysis from real data, every match, every week.

The problem

EPL analysis is mostly noise.

Pundit opinions shift mid-week. Form tables miss context. Head-to-head records don't account for squad changes. xG articles are buried behind paywalls. And when you actually want to look something up — fixture history, a team's away record in top-six games — you're digging through five different sites with no consistent signal.

Pundit takes

Confident opinions with zero accountability and no historical grounding.

Scattered data

Stats live across five tabs. Context is never in the same place as the number.

No time

10 GW fixtures every weekend. You can't do proper research on all of them.

What Castle does differently

One place. Every signal.

We pull match data, run the models, store the recaps, and surface everything through a single interface — no tabs, no paywalls, no noise.

ML Win Probabilities

A trained model outputs 1X2 probabilities for every upcoming fixture — calibrated against historical EPL outcomes, not raw league position.

H 46%  ·  D 27%  ·  A 27%
Projected xG

Expected goals give you a quality-adjusted picture of attacking threat — more reliable than raw goals for spotting mis-priced form.

xG home 1.6  ·  xG away 1.1
RAG-Powered Chat

Ask anything. The AI searches your stored match recaps, stats, and analysis to give you answers grounded in real data — not general football knowledge.

"How has Arsenal performed away in top-six games since GW10?"
Historical Context

Seasons of EPL match data stored and queryable. Patterns across years — not just current form — that tell you whether this fixture type historically delivers goals, cards, or surprises.

Match Analysis Reports

AI-generated pre-match breakdowns covering team form, key matchup risks, tactical angles, and value signals — structured, consistent, fast.

Value Signals

The model flags fixtures where the data meaningfully diverges from public expectation — giving you an edge before kickoff, not a post-match explanation.

How it works

Data in. Clarity out.

1
EPL data is ingested and stored

Match results, stats, and recaps are pulled each gameweek and saved to the database with embeddings for semantic search.

2
Models run pre-match

Scikit-learn models trained on historical EPL data generate 1X2 probabilities and projected xG for upcoming fixtures.

3
AI analysis is generated

An LLM with access to your match database writes structured pre-match reports — grounded in stored data, not hallucination.

4
You ask, it answers

The RAG chat lets you query anything across all stored matches, recaps, and stats — in plain English, in seconds.

Who it's for

If you care about EPL and want to go deeper than the highlights, this is built for you.

Fantasy managers

Fixture difficulty, underlying performance, xG trends — the signals that actually drive FPL points.

Data-curious fans

Get beyond the scoreline. Understand why a team won, what the xG said, and what to expect next.

Informed bettors

Model probabilities and value signals to identify where the data diverges from market expectation. For informational purposes only.

Ready to read the game?

Plans start at $3 / month. No commitment, cancel any time.

For information and entertainment only. No guarantees of financial outcome.