🤖 Autonomous Data Science Agent

An AI-powered data science agent that automatically analyzes any CSV dataset using Claude AI + Machine Learning. Upload a dataset, ask a question in plain English, and the agent decides which tools to use, runs the analysis, trains ML models, and delivers insights — all autonomously.

---

🎬 Demo

---

🧠 How It Works

This project is built on the Model Context Protocol (MCP) architecture:

User (Natural Language)
        ↓
  Streamlit Web UI
        ↓
  Claude AI (Brain)
        ↓
  MCP Tool Router
        ↓
┌───────────────────────────────────┐
│  describe_dataset                 │
│  run_python_analysis              │
│  generate_chart                   │
│  train_ml_model                   │
│  feature_importance               │
└───────────────────────────────────┘

Claude acts as the reasoning engine — it reads your question, decides which tools to call, chains them together intelligently, and explains the results in simple business terms.

---

✨ Features

• Automated Data Cleaning — Auto-fixes missing values, removes duplicates, detects outliers and standardizes column names before analysis
• Intelligent Tool Routing — Claude only calls the tools needed for your question, not all of them
• Statistical Analysis — Correlations, distributions, and relationships between variables
• Auto ML Training — Trains a Random Forest model and returns R² score and error metrics
• Feature Importance — Ranks which variables drive your target metric the most
• Chart Generation — Scatter plots, bar charts, and correlation heatmaps on demand
• Session Isolation — Each session uses its own data, no cross-session chart mixing
• Rate Limit Handling — Friendly messages instead of raw error traces
• API Key Input — Paste your Anthropic key directly in the UI, no .env needed

---

🛠️ Tech Stack

Layer	Technology
AI Brain	Claude (Anthropic)
Agent Protocol	MCP (Model Context Protocol)
ML Engine	Scikit-learn (Random Forest)
Data Processing	Pandas, NumPy
Visualization	Matplotlib, Seaborn
Web UI	Streamlit
Language	Python

---

📁 Project Structure

data-science-agent/
├── app.py                  # Streamlit web UI + agentic loop
├── requirements.txt        # Python dependencies
├── .env                    # Your API key (not committed)
├── .gitignore
├── demo.gif                # Demo recording
├── mcp_server/
│   └── server.py           # MCP server with all 5 tools
├── uploads/                # Uploaded CSV files (not committed)
└── outputs/                # Generated charts (not committed)

---

🚀 Getting Started

1. Clone the repo

git clone https://github.com/yourusername/data-science-agent.git
cd data-science-agent

2. Create and activate virtual environment

python -m venv venv

# Windows
venv\Scripts\activate

# Mac/Linux
source venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. Get your Anthropic API key

👉 https://console.anthropic.com → API Keys → Create Key

5. Run the app

streamlit run app.py

6. Open in browser

http://localhost:8501

Paste your Anthropic API key in the sidebar, upload any CSV, and start asking questions!

---

💬 Example Questions

Question	Tools Used
What columns are in this dataset?	describe_dataset
What drives revenue the most?	run_python_analysis
Show me a correlation heatmap	generate_chart
Train a model to predict revenue	train_ml_model
Which features matter most?	feature_importance
Give me a full analysis	All tools

---

🔧 MCP Tools

describe_dataset

Inspects the CSV and returns shape, column names, data types, missing values, and sample rows.

run_python_analysis

Computes correlations between all numeric columns and the target variable. Labels each as strong/moderate/weak positive or negative.

generate_chart

Generates scatter plots, bar charts, or correlation heatmaps. Returns base64-encoded image — no disk dependency.

train_ml_model

Trains a Random Forest Regressor with an 80/20 train-test split. Returns R² score, MAE, and model quality rating.

feature_importance

Uses Random Forest feature importances to rank which variables drive the target metric. Labels each as High / Medium / Low impact.

---

📊 Sample Dataset

A sample dataset is included to test the agent:

month, marketing_spend, store_count, avg_price, revenue
Jan, 5000, 10, 25.5, 32000
...

Try asking: Give me a full analysis with revenue as the target.

---

🔑 Environment Variables

Create a .env file in the root folder (optional — you can also paste the key in the UI):

ANTHROPIC_API_KEY=sk-ant-your-key-here

---

🔮 Roadmap

• Automated Data Cleaning — auto-detect and fix missing values, duplicates, outliers and column formatting before analysis
• Support for larger real-world datasets (sales, healthcare, finance)
• Time series analysis
• Multi-file upload and joins
• Auto-generated PDF report of full analysis

---

📄 License

MIT License — feel free to use, modify, and share.

---

🙋 Author

Built by Kunal — connecting LLMs with ML automation to build intelligent data tools.