AutoAnalyst: Multi-Agent Data Science Assistant

An intelligent multi-agent system that autonomously performs end-to-end data analysis, ML model training, and report generation — powered by CrewAI and GPT-4.

---

What It Does

You upload any CSV file. The system automatically:

• Analyzes the data (EDA, statistics, correlations)
• Trains and compares multiple ML models
• Generates interactive visualizations
• Produces a full written report

No manual steps. No code required from the end user.

---

System Architecture

The system is organized into 5 phases, each adding a new layer of capability:

Phase 1 — Data Analysis Pipeline 5 specialized AI agents run sequentially, each passing output to the next:

1. Data Loader Agent — validates data quality and structure
2. EDA Specialist Agent — performs statistical analysis
3. Visualization Expert Agent — generates and interprets charts
4. Insights Analyst Agent — identifies patterns and recommendations
5. Report Writer Agent — produces a professional markdown report

Phase 2 — Machine Learning Pipeline 4 ML agents handle automated modeling:

• Feature Engineer Agent — selects and transforms features
• Model Selector Agent — chooses appropriate algorithms
• Model Trainer Agent — trains Logistic Regression, Decision Tree, Random Forest, Gradient Boosting
• Model Evaluator Agent — compares models on Accuracy, Precision, Recall, F1, ROC-AUC

Phase 3 — Streamlit Web Dashboard Interactive UI built with Streamlit:

• Drag-and-drop CSV upload
• Real-time data exploration with Plotly charts
• One-click ML model training with live progress
• Results dashboard with model comparison and feature importance

Phase 4 — Advanced ML

• AutoML with Optuna for hyperparameter tuning — achieved 37.1% F1 improvement (0.30 to 0.41)
• SHAP explainability — shows why the model made each prediction
• Ensemble methods — Voting and Stacking classifiers

Phase 5 — Production API

• FastAPI REST API deployed on Render.com
• Docker containerized for consistent deployment
• Swagger UI documentation at /docs
• Response time under 200ms
• Supports single prediction and batch CSV prediction

---

ML Results (Healthcare Dataset)

Trained and evaluated on 500 patient records predicting heart disease:

| Model | Accuracy | F1-Score | ROC-AUC | |---------------------|----------|----------|---------| | Logistic Regression | 0.75 | 0.67 | 0.82 | | Decision Tree | 0.72 | 0.64 | 0.78 | | Random Forest | 0.81 | 0.75 | 0.88 | | Gradient Boosting | 0.79 | 0.72 | 0.86 |

Best Model: Random Forest — auto-selected based on F1-Score

Top Predictors (via SHAP): Cholesterol, Age, Blood Pressure

AutoML Result: Random Forest F1 improved from 0.30 to 0.41 (+37.1%) using Optuna

---

API Endpoints

| Method | Endpoint | Description | |--------|-----------------|------------------------------------| | GET | / | Welcome message and API info | | GET | /health | Health check and model status | | POST | /predict | Single patient prediction | | POST | /batch-predict | Batch predictions from CSV | | GET | /stats | Usage statistics | | GET | /model-info | Model details and features | | GET | /docs | Interactive Swagger UI |

Live API: https://autoanalyst-api.onrender.com

---

Tech Stack

| Layer | Technology | |----------------|-------------------------------------------------| | AI Agents | CrewAI | | Language Model | OpenAI GPT-4o-mini | | ML | Scikit-learn, XGBoost, LightGBM | | AutoML | Optuna | | Explainability | SHAP | | API | FastAPI | | Frontend | Streamlit | | Visualization | Plotly, Matplotlib, Seaborn | | Data | Pandas, NumPy | | Deployment | Docker, Docker Compose, Render.com | | CI/CD | GitHub Actions | | Language | Python 3.12 |

---

Screenshots

Web Dashboard

Home Page

ML Training in Progress

Model Performance Table

Results & Insights

---

Data Analysis Outputs

Distribution Analysis

Correlation Heatmap

Categorical Analysis

---

ML Results

Model Performance Comparison

Confusion Matrices

ROC Curves

Feature Importance (Random Forest)

Full Analysis Collage

---

Quick Start

Prerequisites

• Python 3.8+
• OpenAI API key

Installation

git clone https://github.com/Sakshi3027/AutoAnalyst.git
cd AutoAnalyst
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
# Add your OPENAI_API_KEY to .env

Run the Web Dashboard (Recommended)

streamlit run streamlit_app.py

Opens at http://localhost:8501

Steps:

1. Click "Load Sample Data" in the sidebar
2. Go to "Data Analysis" to explore
3. Go to "ML Pipeline" and click "Train Models"
4. View "Results" for full analysis

Run Data Analysis Only

python main.py

Output: outputs/analysis_report.md

Run ML Pipeline Only

python main_ml.py

Output: outputs/ml_analysis_report.md and outputs/best_model.pkl

Run Advanced ML (AutoML + SHAP)

python main_advanced.py

---

Docker Deployment

# Build
docker build -t autoanalyst .

# Run
docker run -p 8000:8000 --env-file .env autoanalyst

# Or with docker-compose
docker-compose up -d

---

Use Saved Model

import joblib
import pandas as pd

model = joblib.load('outputs/best_model.pkl')

new_patient = pd.DataFrame({
    'age': [55],
    'bmi': [28.5],
    'blood_pressure_systolic': [140],
    'blood_pressure_diastolic': [90],
    'cholesterol': [220],
    'glucose': [120],
    'exercise_hours_per_week': [3],
    'gender_Male': [1],
    'smoker_Yes': [1]
})

prediction = model.predict(new_patient)
probability = model.predict_proba(new_patient)

print(f"Prediction: {'Heart Disease' if prediction[0] == 1 else 'No Heart Disease'}")
print(f"Probability: {probability[0][1]:.2%}")

---

Project Structure

AI_agents/
├── agents/                        # 12 AI agent definitions
│   ├── data_loader_agent.py
│   ├── eda_agent.py
│   ├── visualization_agent.py
│   ├── insight_agent.py
│   ├── report_agent.py
│   ├── feature_engineer_agent.py
│   ├── model_selector_agent.py
│   ├── model_trainer_agent.py
│   ├── model_evaluator_agent.py
│   ├── automl_agent.py
│   ├── explainability_agent.py
│   └── deep_learning_agent.py
├── utils/
│   ├── data_utils.py              # EDA and visualization helpers
│   └── ml_utils.py                # ML training and evaluation
├── data/
│   └── healthcare_data.csv        # Sample dataset (500 records)
├── outputs/                       # Generated reports and models
├── screenshots/                   # UI screenshots
├── api.py                         # FastAPI REST API
├── Dockerfile
├── docker-compose.yml
├── main.py                        # Phase 1 pipeline
├── main_ml.py                     # Phase 2 ML pipeline
├── main_advanced.py               # Phase 4 advanced ML
├── streamlit_app.py               # Phase 3 web dashboard
├── requirements.txt
└── README.md

---

Key Achievements

• 12 AI agents for fully automated data science pipeline
• 37.1% F1-Score improvement through AutoML (Optuna)
• Production API deployed on Render.com with under 200ms response time
• SHAP explainability for model transparency
• Full Docker containerization
• End-to-end: raw CSV in, trained model + report out

---

Use Cases

• Healthcare data analysis and prediction
• Financial report generation
• Marketing analytics
• Research data exploration
• Enterprise data science automation

---

Author

Sakshi Chavan

• GitHub: github.com/Sakshi3027
• Live API: autoanalyst-api.onrender.com

---

License

MIT License