Agentic AI · Full-Stack · Claude

Every FPL decision,
answered at your fingertips.

Eleven million people play Fantasy Premier League, and every week each one faces the same questions: who to transfer, who to captain, when to play a chip. Pocket Scout reads your actual squad and answers all of them — personalized advice for your team, your budget, your gameweek. And because every call is grounded in real FPL data, it's advice you can actually trust.

Deterministic scoring engine Grounded via tool calls Backtested · 10 seasons Vitest · 331 tests Anthropic Claude Public FPL API Full-stack TypeScript
11M+
FPL managers
facing the same calls
0.53
Model rank correlation
(vs FPL's own 0.59)
10
Seasons of data
backtested
0
Invented stats —
grounded in real data
Getting started

Two ways in — your team, or a test drive.

The journey starts one of two ways. Enter your FPL manager ID and Pocket Scout reads your real squad — your players, your bank, the exact free transfers you're holding. Or click "Explore without a team" and it builds a sample dream XV from the season's best returns, so you can try the whole thing before you ever type in an ID.

Pocket Scout entry screen with a field to enter your FPL manager ID and an option to explore without a team
Enter a manager ID to get scoutedYour real team
Demo mode showing the Scout's dream XV built from the season's best returns, with ratings and an Ask the Scout briefing
…or explore a ready-made dream XVNo ID needed

The moment you're in

Your whole gameweek, on one screen.

The pitch paints instantly: every player rated 0–10 on a position-aware scale, and a verdict banner across the top calling this week's move — the transfer to make, who to captain, and whether to play a chip. No spreadsheets, no twelve open tabs — just your team and what to do with it.

Full walkthrough — rate your squad, plan transfers, ask the Scout▶ 2-min demo

The plan

Transfers, captaincy, and chips — sorted to your budget.

Open the breakdown and the week's hardest calls become a clear plan across three horizons: an immediate This Week move, a multi-gameweek Long Term outlook, and a season-long Chips strategy. It plans inside your real budget — tell it how many free transfers you're holding (0–5) and it works within that, stacking moves or banking one rather than burning it on a marginal swap. Every move is judged in expected points, so when nothing clears the bar it tells you to hold — the opposite of the churn most tools push.

The planner drawer showing This Week, Long Term and Chips tabs with a high-confidence transfer plan, captaincy pick and chip recommendation
This Week / Long Term / Chips, with confidence and net-points deltasPersonalized to your squad
Player detail dialog for Bruno Fernandes showing form, minutes, expected points and a link to the official Premier League page
Tap any player for form, minutes, and expected pointsPL deep link

Ask the Scout

Then ask it anything.

"Should I captain Haaland this week?" "Who's a rotation risk before the run-in?" "Is it worth taking a hit?" The questions managers stew over all week get an instant, specific answer — a chat that knows your actual squad and the plan it just built, and talks you through the call like a mate who's done the homework.

Ask the Scout answering a rotation-risk question with player-specific expected points and reasoning
Ask anything about your team — reasoned, squad-specific answersInstant, specific

Why you can trust it

Advice that holds up when you check it.

Managers check the numbers afterward, so the hard figures are computed before any language model is involved: a deterministic base phase rates your squad and proposes moves straight from the FPL API, and only then does Claude reason over those numbers to explain the call. The model is the pundit; it never invents the underlying stats.

📊
Base phase — deterministic
Rates every player 0–10 and proposes transfers/captain instantly from FPL data. No LLM in the loop.
🧠
Insights phase — Claude
Reasons over the computed numbers to explain the advice, cached for speed and cost.
Personalized verdict
A single, actionable call on transfers, captaincy, and chips — tailored to your squad and bank.
Grounded, not guessing

Most FPL chatbots happily hallucinate prices, form, and fixtures. Pocket Scout fetches real stats through tool calls before it answers — and if the evidence isn't there, it says so instead of making one up. It also treats every message as a question to answer, not an instruction to obey, so it can't be talked off its own plan.

Ask the Scout declining an off-topic request and steering the conversation back to the user's FPL squad
Try to take it off-topic — or inject it — and it stays strictly FPLCan't go off-script

Measure, don't assume

A real app, with decisions earned by data.

Pocket Scout is a deployed full-stack TypeScript app — not a notebook demo — guarded by a 331-test suite so behaviour stays pinned as the model and rules evolve. And it was built eval-first: a point-in-time backtest over ten seasons of FPL data drove the model, and the calls that shaped it were made on evidence, including the honest negatives.

0.53
Rank corr.
Fitting the player-ranking model on ten seasons of history lifted within-position rank correlation from ~0.33 to ~0.53 — closing most of the gap to FPL's own ~0.59 benchmark.
Hold
gate added
Squad replays showed the optimizer over-recommended transfers. An evidence-based, points-based hold gate now banks a free transfer when moving isn't worth it.
Rejected
A proposed fixture-difficulty model upgrade looked sensible but failed its backtest — so it was rejected. Convention lost to the numbers.

Personal reflection
Behind the build

I built Pocket Scout because I'm one of those eleven million managers, and the weekly decisions — who to bring in, who to captain, when to burn a chip — are genuinely hard and genuinely fun to argue about. I wanted a smart, opinionated advisor that actually knows my team, so "what should I do this week?" is one question away.

The part I'm proudest of is letting evaluation overrule intuition. More than once the data killed a change that felt obviously right — and the eval harness got the final word, not me. That's the same bar I hold every project on this site to.

See it live — or dig into the build.