🚀 AI Agent Hackathon Platform 🎯 Objective
Welcome to the AI Agent Hackathon! -In this challenge, you will: -Explore a real-world dataset -Build features based on your understanding -Train a machine learning model -Create your own AI Agent (Streamlit app) -Submit predictions -Get evaluated automatically using metrics + AI judge
🧠 Architecture Overview Dataset → Exploration → Feature Engineering → Model Training → Predictions ↓ AI Agent (Streamlit) ↓ Evaluation System ↓ Leaderboard 🏗️ Project Structure
ai-agent-hackathon/ │ ├── data/ # Dataset (already provided) ├── notebooks/ # Step-by-step notebooks ├── models/ # Saved model (model.pkl) ├── outputs/ # Your submission goes here ├── evaluations/ # Evaluation system ├── app/ # Streamlit UI (Agent + Leaderboard) ├── README.md
🧭 PHASE-WISE GUIDE 🔹 PHASE 1 — Data Exploration
📘 Notebook: 01_data_exploration.ipynb
What you will do:
Understand dataset structure Identify patterns Analyze relationships Choose your target variable
🎯 Target Options:
target_churn → Classification target_fraud → Classification target_revenue → Regression ** 🔹 PHASE 2 — Feature Engineering**
📘 Notebook: 02_feature_engineering.ipynb
What you will do:
Clean data Encode categorical variables Create new features 💡 Examples: activity_score engagement_ratio spend_per_transaction
👉 This is the most important phase 👉 Better features = better model
🔹 PHASE 3 — Model Training
📘 Notebook: 03_model_training.ipynb
What you will do:
Train ML models Compare performance Save best model
🤖 Models you can use:
Type Models Basic Logistic Regression Tree-based Random Forest, Gradient Boosting Advanced XGBoost Regression Linear Regression
🔹 PHASE 4 — Generate Predictions
📘 Notebook: 04_generate_predictions.ipynb
What you will do: Load model Predict on test data Save output
📤 FINAL OUTPUT FORMAT actual,prediction 📁 Save your file as: outputs/YOURNAME_predictions.csv
Example:
outputs/harshit_predictions.csv 🔹 PHASE 5 — Build AI Agent
📘 File: app/app.py
Run your agent: streamlit run app.py What your agent does:
Takes user input Uses your trained model Shows predictions Displays confidence
🔹 PHASE 6 — Evaluation
📘 File: app/leaderboard.py
Run leaderboard: streamlit run leaderboard.py Click:
👉 Run Evaluation
🏆 Scoring System For Classification:
Accuracy → 50% F1 Score → 30% AI Judge → 20% For Regression: MSE (lower better) → 60% R² Score → 20% AI Judge → 20%
🤖 AI JUDGE (SPECIAL FEATURE)
Your model is also evaluated by AI based on:
Prediction quality Generalization Business usability
🏆 LEADERBOARD
Automatically updated Supports tie-breaking Fair ranking system
✅ Use any model ✅ Create your own features ✅ Customize your AI agent
❌ Not Allowed:
Changing output format Incorrect file naming Multiple submissions after deadline
🧠 HOW TO USE CHATGPT (VERY IMPORTANT)
You are encouraged to use ChatGPT to improve your solution.
🔥 Example Prompts Feature Engineering: Suggest 5 advanced features for churn prediction dataset Model Selection: Which model is best for classification with tabular data? Hyperparameter Tuning: How to improve RandomForest performance? Debugging: Why is my model overfitting? Evaluation: How to improve F1 score? 💡 Pro Tips
Focus on feature engineering first
Try multiple models Check feature importance Keep model simple but effective
**🎯 FINAL CHECKLIST Before submission:
✔ Model trained ✔ Predictions generated ✔ File saved in outputs/ ✔ File name correct ✔ Columns correct
🚀 FINAL GOAL
Build an AI Agent that is:
✅ Accurate ✅ Intelligent ✅ User-friendly ✅ Business-ready
**🔥 GOOD LUCK
“Your model is your brain, but your agent is your product.”
🚀 Go build something amazing!**