⚽ V-TAC: Vision Tactical AI Coach

🌟 Overview

V-TAC (Vision Tactical AI Coach) is a next-generation football AI system that offers accurate pre-match predictions, live tactical decisions, and fatigue-aware substitution advice. Designed for analysts and coaches, V-TAC combines machine learning, real-time API data, and a chatbot interface.

---

⚖️ Data Collection & Preparation

✅ Data Sources

• Match Data: 70,000+ games from API-Football
• Player Data: 120,000+ player records with live stats (passes, tackles, fatigue)
• Leagues: Premier League, La Liga, Bundesliga, Serie A, Saudi Pro League, etc.

⚙️ Preprocessing Pipeline

• Merging data across seasons
• Cleaning missing values
• Feature engineering (recent form, H2H, xG, formation success)
• Encoding teams/formations
• Scaling using StandardScaler

---

🤖 Core AI Models

1️⃣ Pre-Match Outcome Prediction

• Goal: Predict Win/Draw/Loss before kickoff
• Model: AutoGluon.Tabular
• Type: Multiclass classification
• Features: team rank, points, pass %, goals, form, formation, H2H
• Accuracy: ~77% after oversampling

---

2️⃣ Live Match Outcome Prediction (Every 5 Minutes)

• Goal: Update predictions live using real-time stats
• Model: AutoGluon + real-time snapshot
• Data: goals, possession, shots, pass %, formation, momentum
• Metric: Log loss < 0.27

Example:

Chelsea → Win: 63% | Draw: 25% | Loss: 12%
Recommendation: ✅ Keep the formation, momentum is high.

---

3️⃣ Player Fatigue Estimator

• Goal: Detect when a player is too tired
• Inputs: minutes, passes, duels, distance covered
• Output: fatigue score (0 to 1)
• Frequency: every 5 minutes for each player
• Accuracy: ~97%

---

🧭 Recommendation & Interface Components

4️⃣ Formation Suggestion Engine

• Type: Rule-based
• Recommends the best formation historically vs opponent

Best Formation: 4-2-3-1
→ vs 4-3-3 = 70% win rate

---

5️⃣ AI Chatbot (LLM-powered)

• Based on OpenChat
• Answers tactical questions using real predictions
• Live updates trigger smart language output

"Switch to 4-4-2, midfield fatigue is increasing."

---

6️⃣ CSV Agent (LangChain)

• Used for QA over match/player CSVs
• Helpful during model validation/debugging

---

7️⃣ 3D Avatar + Edge-TTS + Rhubarb Lip Sync

• TTS: Converts predictions to voice
• Lip Sync: Generates visemes using Rhubarb
• 3D Avatar: Visual face for match recommendations

---

🔁 Real-Time Architecture Flow

video_2025-04-25_10-29-07.mp4

• ⏱ Updates every 5 mins during match
• 📥 Sends outputs to Chatbot / Avatar
• 🔁 After match: update training data + fine-tune model

---

💻 UI/UX (Figma Preview)

🔗 View Prototype

---

🧠 AI Tech Stack

Layer	Technology
Model Training	AutoGluon, Scikit-learn
Real-Time Updates	API-Football + custom logic
Data Interaction	LangChain CSV Agent
Voice Interaction	Edge-TTS + Rhubarb
Avatar Interface	React + Flask
Language Model	OpenChat

---

👥 Team

Name	Role
Abdulrahman AlNashri	AI Modeling + UI/UX
Osama AlGhamdi	AI Modeling + Sport
Rawaa AlTurkistani	Security + Presentations
Ghadir Najm	Business + Coordination

---

🚀 What's Next?

• 🎮 GPS/IMU sensors for live fatigue
• 📊 Deep learning for player embeddings
• 🧠 Fully autonomous LLM assistant
• ⚽ Deployment in real matches with clubs