---

🤔 What if you could build an entire Data Science ecosystem from scratch?

No Kaggle datasets. No YouTube tutorials. No shortcuts.

Just raw Python, real business logic, and 12 layers of pure Data Science.

That's CHANAKYA — a 100% original end-to-end AI Commerce Intelligence Platform built on a synthetically generated Indian e-commerce dataset with real brands, real cities, real business patterns.

💡 "Most people download a dataset. I built one."

---

📊 The Numbers Speak

🛒 Orders	👥 Customers	💰 Revenue	📦 Products	🏙️ Cities
5,010	1,000	₹2.67 Crore	50 Indian Brands	28 Cities

---

🏗️ 12 Layers of Pure Data Science

┌─────────────────────────────────────────────────────────────┐
│                     CHANAKYA ECOSYSTEM                       │
├──────┬──────────────────────────┬───────────────────────────┤
│ L1   │ Data Source + Simulation │ 5,010 live orders         │
│ L2   │ ETL Pipeline             │ Production-grade logging  │
│ L3   │ SQL Data Warehouse       │ Star Schema + 7 queries   │
│ L4   │ EDA                      │ 16 professional charts    │
│ L5   │ RFM Segmentation         │ 9 segments + animation 🎬 │
│ L6   │ Anomaly Detection        │ 34 frauds caught          │
│ L7   │ ML Demand Forecasting    │ Gradient Boosting wins    │
│ L8   │ Deep Learning LSTM       │ Time series forecasting   │
│ L9   │ Churn Prediction         │ 477 at-risk customers     │
│ L10  │ Power BI Dashboard       │ Dark theme + DAX measures │
│ L11  │ ARTHA Agentic AI         │ LLaMA 3.3 70B via Groq    │
│ L12  │ Streamlit Deployment     │ Live on the internet      │
└──────┴──────────────────────────┴───────────────────────────┘

---

🖥️ Streamlit Live App

---

🤖 ARTHA — Agentic AI with Auto Visualizations

Named after Kautilya's Arthashastra — the ancient treatise on wealth and governance

ARTHA doesn't just answer questions. It thinks, analyzes, and visualizes — all in real time.

---

📊 Power BI Executive Dashboard

Dark theme | 6 KPIs | 5 Visuals | 7 DAX Measures | 3 Interactive Slicers

---

📈 Layer 4 — Exploratory Data Analysis

16 professional charts revealing real business insights

---

🎬 Layer 5 — RFM Customer Segmentation

9 customer segments | Animated visualization

---

🚨 Layer 6 — Anomaly Detection

34 fraud accounts caught using Z-Score + Isolation Forest

---

🤖 Layer 7 — ML Demand Forecasting

4 models compared | Gradient Boosting wins with R2: 0.2891

Linear Regression    → R2: 0.2652
Decision Tree        → R2: 0.0780
Random Forest        → R2: 0.2522
Gradient Boosting    → R2: 0.2891 ✅ WINNER

📝 R2 scores are moderate due to limited monthly data (37 points). In production with daily transaction data, accuracy will significantly improve. This demonstrates honest ML evaluation — not overfitting to small datasets.

---

🧠 Layer 8 — Deep Learning LSTM

Time series revenue forecasting with TensorFlow

---

📉 Layer 9 — Churn Prediction

477 at-risk customers identified | Data leakage detected & fixed

v1 with days_since_last  → 100% accuracy ❌ (LEAKAGE!)
v2 without leakage       → 67% accuracy  ✅ (HONEST)

---

💡 Key Business Insights

📈  Electronics drives 75% revenue — but has the LOWEST profit margin
🏆  129 Champion customers generate ₹83.4L — top 13% = 31% revenue
⚠️  34 fraud accounts detected using Z-Score + Isolation Forest
📉  477 customers predicted to churn — before they actually left
🎯  November is peak month — festive season spike clearly visible
💳  UPI dominates at 34.9% — Digital India is real

---

🔬 Technical Depth

SQL — 8 Concepts Used

WITH customer_ltv AS (
    SELECT customer_id, SUM(revenue) as lifetime_value,
    RANK() OVER (ORDER BY SUM(revenue) DESC) as ltv_rank
    FROM fact_order_items GROUP BY customer_id
)
SELECT * FROM customer_ltv ORDER BY lifetime_value DESC;

Skills Demonstrated

Data Engineering	Machine Learning	Business Analytics
✅ ETL Pipeline	✅ 4 ML Models Compared	✅ RFM Segmentation
✅ Star Schema Design	✅ Deep Learning LSTM	✅ Anomaly Detection
✅ Feature Engineering	✅ Churn Prediction	✅ Power BI DAX
✅ Data Validation	✅ Data Leakage Detection	✅ Agentic AI

---

🚀 Tech Stack

Language      : Python 3.11
Database      : MySQL (Star Schema)
ML Libraries  : Scikit-learn, TensorFlow, Keras
Data          : Pandas, NumPy, Faker (Indian locale)
Visualization : Matplotlib, Seaborn, Plotly, Power BI
AI Model      : LLaMA 3.3 70B via Groq API
Frontend      : Streamlit
Deployment    : Streamlit Cloud
Security      : python-dotenv
Version Ctrl  : Git + GitHub

---

⚡ Run Locally

git clone https://github.com/DeveshShukla23/CHANAKYA.git
cd CHANAKYA
pip install -r layer12_streamlit_api/requirements.txt
echo "GROQ_API_KEY=your_groq_key_here" > layer12_streamlit_api/.env
cd layer12_streamlit_api
streamlit run app.py

---

👨‍💻 Author

Devesh Shukla Data Analyst | Data Scientist | Builder

6 months internship experience | Passionate about turning data into decisions

"Data is the new Arthashastra. CHANAKYA masters both." 🏛️

⭐ Star this repo if you found it useful! ⭐