Football Assistant Coach - AI Player Selection for Your Squad

📋 Project Overview

Challenge: Selecting the starting lineup for a 22-player squad is one of the hardest decisions a coach makes each week. Form fluctuates. Training observations are subjective. And traditional coaching instincts can favour star players over team players — even when the data says otherwise.

Solution: Football Assistant Coach runs four AI-evaluated training sessions every week, scores each player on team chemistry (max 50), individual performance (max 35), and universal metrics (max 15), then recommends the optimal Starting XI, bench, and identifies players at transfer risk — all using a formation-aware selection engine.

Key Design Decisions

Team chemistry outweighs individual talent — max 50 pts deliberately rewards players who press together, track back, make support runs, and communicate. Star players who don't contribute to the team shape score lower.
Formation-aware selection — the app doesn't just rank 11 players; it fills GK, DEF, MID, FWD quotas according to the chosen formation (4-3-3, 4-4-2, 3-5-2).
Hybrid advisor — coach declares a preferred formation, but the AI surfaces a warning and recommends an alternative if this week's in-form players better suit a different shape.
Rolling form — weekly score = this week (60%) + last week (40%), smoothing out one-off bad sessions.

📊 Scoring System

50

Team Chemistry (max)

35

Individual Performance (max)

15

Universal Metrics (max)

100

Total Score

Selection Thresholds

Form Score	Status
65 and above	Starting XI eligible
50 – 64	Bench eligible
40 – 49	Unused — needs more training
Below 40 (2 weeks)	Transfer shortlist

Training Session Weights

Day	Session	Weight	Focus
Monday	Fitness & Conditioning	15%	Universal 60%, Individual 30%, Chemistry 10%
Wednesday	Position-Specific Drill	25%	Individual 70%, Universal 20%, Chemistry 10%
Thursday	Tactical Shape & Pressing	25%	Chemistry 70%, Individual 10%, Universal 20%
Friday	Full Team Scrimmage	35%	Equal thirds

⚙️ How It Works

22-Player Squad

2 GK, 6 DEF, 8 MID, 6 FWD — each with name, age, style, strength, weakness, and base form built in. No external dataset required.

Position-Specific AI Coach

Claude Haiku takes on the role of a goalkeeping coach, defensive coach, midfield coach, or attacking coach — matching the persona to each player's position.

4 Sessions Per Week

Each player is evaluated in Monday fitness, Wednesday drills, Thursday tactics, and Friday scrimmage. Each session has different scoring weights.

Formation Advisor

After simulation, the app scores all three formations against the week's in-form players and warns you if your preferred formation isn't the best fit.

Transfer Risk Detection

Players who score below 40 for two consecutive weeks are automatically flagged on the transfer shortlist — a pattern-based warning, not a one-off bad session.

Bench Coverage Rule

Bench always includes 1 GK before filling the remaining 4 spots. Ensures position coverage for substitutions in any match scenario.

🛠️ Tech Stack

Python 3 Claude Haiku (Anthropic) Streamlit Plotly Pandas python-dotenv Rich (CLI) Anthropic SDK

Architecture

config.py — all constants: scoring caps, session weights, formation quotas, thresholds
players.py — 22 hardcoded player profiles with attributes
coach.py — PlayerAgent: calls Claude with position-specific system prompt, returns structured JSON scores
simulation.py — runs 4 sessions per week, computes weighted weekly score and rolling form
report.py — formation-aware selection engine + fit scoring for hybrid advisor
app.py — Streamlit UI with dark theme, squad cards, bar chart, formation advisor

Sample Claude Output (per player per session)

{
  "individual_score": 28,
  "chemistry_score": 41,
  "universal_score": 12,
  "highlight": "Excellent pressing coordination during shape drills.",
  "concern": "Needs to improve distribution under pressure."
}

🚀 Setup & Run

pip install -r requirements.txt

Add your Anthropic API key to the .env file, then:

# Double-click run.bat (Windows)
# or run manually:
streamlit run app.py

The app opens at http://localhost:8501. Pick formation → click Run Training Week → review selection.

⚽ Football Assistant Coach