FastAPI ML API - Sentiment Analysis

🎯 Overview

A production-ready FastAPI boilerplate that combines robust user management with Machine Learning capabilities. Built with a clean architecture, JWT authentication, SQL Server persistence, and a trained sentiment analysis model.

Perfect foundation for building AI-powered applications that require user authentication, data persistence, and intelligent text analysis.

✨ Features

🔐 Security & Authentication

• JWT Token-based authentication with HTTPBearer
• Password hashing using bcrypt
• Protected routes with dependency injection
• Environment-based configuration

🗄️ Database

• SQL Server integration via SQLAlchemy + pyodbc
• Stored procedures for optimized queries
• Connection pooling for performance
• Clean repository pattern

🤖 Machine Learning

• Sentiment Analysis (Spanish language)
• Naive Bayes + TF-IDF classifier
• Real-time predictions via REST API
• Batch processing support
• 95%+ accuracy on validation set

🏗️ Architecture

• Layered architecture: Clean separation of concerns
• Dependency injection: Flexible and testable
• Pydantic schemas: Type-safe data validation
• Modular design: Easy to extend and maintain

🚀 Quick Start

Prerequisites

• Python 3.10+
• SQL Server (local or remote)
• ODBC Driver 17/18 for SQL Server

Installation

1. Clone the repository

git clone https://github.com/javsan77/FastAPI-ML-API---Sentiment-Analysis.git
cd FastAPI-ML-API---Sentiment-Analysis

1. Create virtual environment

python -m venv venv
source venv/bin/activate

1. Install dependencies

pip install -r requirements.txt

1. Configure environment variables

Create a .env file:

DB_SERVER=localhost
DB_NAME=fastapi_user_api
DB_USER=your_username
DB_PASSWORD=your_password
DB_DRIVER=ODBC Driver 18 for SQL Server

1. Initialize database

Execute scripts.sql in your SQL Server instance to create tables and stored procedures.

1. Train the ML model

python train_model_final.py

1. Run the application

uvicorn app.main:app --reload

🎉 API is live at: http://localhost:8000

📚 Interactive docs: http://localhost:8000/docs

📂 Project Structure

fastapi-ml-api/
├── app/
│   ├── config/              # Database, Security, Environment configuration
│   │   ├── database.py
│   │   └── security.py
│   ├── core/                # Core utilities (bcrypt)
│   │   └── security.py
│   ├── dependencies/        # Dependency injection (Auth)
│   │   └── auth_dependency.py
│   ├── ml/                  # 🤖 Machine Learning Module
│   │   ├── models/
│   │   │   └── sentiment_model.py
│   │   └── services/
│   │       ├── text_preprocessor.py
│   │       └── sentiment_service.py
│   ├── repositories/        # Data access layer (Stored Procedures)
│   │   └── user_repository.py
│   ├── routers/             # API endpoints
│   │   ├── auth_router.py
│   │   ├── user_router.py
│   │   └── ml_router.py     # 🤖 ML endpoints
│   ├── schemas/             # Pydantic models
│   │   ├── auth_schema.py
│   │   ├── user_schema.py
│   │   └── ml/
│   │       └── sentiment_schema.py
│   ├── services/            # Business logic
│   │   ├── auth_service.py
│   │   └── user_service.py
│   └── main.py              # Application entry point
├── scripts.sql              # Database setup
├── train_model_final.py     # ML model training script
├── requirements.txt
└── .env                     # Environment variables (gitignored)

📖 API Documentation

Authentication Endpoints

User Endpoints

Machine Learning Endpoints

🤖 Machine Learning

Sentiment Analysis Model

The API includes a trained sentiment analysis model for Spanish text that classifies input as:

• POSITIVE 😊
• NEGATIVE 😞
• NEUTRAL 😐

Model Specifications

• Algorithm: Multinomial Naive Bayes
• Vectorization: TF-IDF with n-grams (1-3)
• Training samples: 200+
• Validation accuracy: 95%+
• Language: Spanish

Example Usage

Single text analysis:

curl -X POST "http://localhost:8000/ml/sentiment/analyze" \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Este producto es increíble, me encanta mucho"
  }'

Response:

{
  "text": "Este producto es increíble, me encanta mucho",
  "sentiment": "POSITIVE",
  "confidence": 0.96,
  "scores": {
    "POSITIVE": 0.96,
    "NEGATIVE": 0.02,
    "NEUTRAL": 0.02
  },
  "analyzed_at": "2026-01-02T10:30:00"
}

Batch analysis:

curl -X POST "http://localhost:8000/ml/sentiment/batch" \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "texts": [
      "Me encanta este servicio",
      "Pésima experiencia",
      "Está bien, nada especial"
    ]
  }'

Retraining the Model

To retrain with your own data, edit train_model_final.py and run:

python train_model_final.py

The model will be saved to app/ml/models/sentiment_classifier.pkl

🔧 Configuration

Environment Variables

Security Settings

Located in app/config/security.py:

SECRET_KEY = "CHANGE_THIS_IN_PRODUCTION"
ALGORITHM = "HS256"
ACCESS_TOKEN_EXPIRE_MINUTES = 60

⚠️ Important: Change SECRET_KEY before deploying to production!

Production Checklist

• Change SECRET_KEY in security.py
• Set strong database password
• Configure CORS policies
• Enable HTTPS/SSL
• Set up logging and monitoring
• Configure rate limiting
• Backup database regularly
• Use production-grade WSGI server (Gunicorn/Uvicorn)

🎓 Use Cases

This boilerplate is ideal for:

• Customer feedback analysis systems
• Social media sentiment monitoring
• Support ticket classification
• Product review analysis
• Chat applications with sentiment detection
• AI-powered chatbots with user management
• NLP research projects with API interface

📝 License

This project is licensed under the MIT License - see the LICENSE file for details.

👤 Author

Javier Sanchez Ayte

• GitHub
• LinkedIn
• Portfolio

🙏 Acknowledgments

• FastAPI - Modern web framework
• scikit-learn - Machine learning library
• SQLAlchemy - SQL toolkit
• Pydantic - Data validation