🧠 Image Prediction API — Technical Documentation

📋 Overview

This project implements a FastAPI-based image prediction service integrated with a PostgreSQL database, Dockerized microservices, and complete CI-level testing. It provides endpoints for user management, authentication, feedback storage, and image classification through a separate ML service.

✅ All unit and integration tests pass successfully (pytest) ✅ Docker & seed scripts verified ✅ Production-ready modular structure

---

🏗️ Architecture Overview

📂 Directory Structure

assignment/
├── app/
│   ├── auth/
│   │   ├── jwt.py
│   │   ├── router.py
│   ├── db.py
│   ├── model/
│   │   ├── router.py
│   │   ├── schema.py
│   │   ├── services.py
│   ├── feedback/
│   │   ├── router.py
│   │   ├── models.py
│   ├── user/
│   │   ├── models.py
│   │   ├── schema.py
│   │   ├── services.py
│   │   ├── validator.py
│   │   ├── router.py
│   ├── utils.py
│   ├── seed.py
│   └── main.py
├── docker-compose.yml
├── Dockerfile
├── requirements.txt
├── Makefile
└── tests/

---

🐳 Dockerized Stack

Service	Description	Container Name
FastAPI backend	Main API app	ml_api
ML service	Model prediction service	ml_service
PostgreSQL	Database	postgres_db
Redis	Task/message queue	assignment-redis-1
UI	Optional dashboard	ml_ui

---

⚙️ Setup & Execution

🧩 Build and Run

make rebuild-api   # rebuild API container
make up            # start all services

🧹 Clean up

make down          # stop and remove containers
make clean         # prune unused images/volumes

🌱 Seed Database

make seed

Creates or updates an admin user:

email: admin@example.com
password: admin

---

🧪 Testing

Run all tests

make test-api-all

Individual tests

make test-api-model     # model endpoint tests
make test-api           # API layer tests
make test-model         # ML service tests

One-shot full verification

make shot-all

Runs full cycle: rebuild → up → seed → test-api → test-model

---

🧰 Key Endpoints

Method	Endpoint	Description
POST	/login	Obtain JWT token
GET	/me	Get current user
POST	/model/predict	Upload image and get prediction
GET	/user/	Get all users
POST	/user/	Create user registration
GET	/user/{id}	Get user by ID
DELETE	/user/{id}	Delete user
GET	/feedback/	Get all feedback entries
POST	/feedback/	Create new feedback

---

🔑 Authentication

• OAuth2 Bearer Token (JWT)

• Use /login to obtain token

• Add header to protected requests:

  Authorization: Bearer <your_token>
  

---

🧬 Model Service

The /model/predict endpoint interacts with a separate ML container. It:

1. Validates image type (PNG/JPG/JPEG/GIF)
2. Computes SHA-256 hash for filename
3. Stores it under /tmp (or configured UPLOAD_FOLDER)
4. Sends it to the ML service for classification
5. Returns:

{
  "success": true,
  "prediction": "cat",
  "score": 0.95,
  "image_file_name": "fakehash123"
}

---

🧑‍💻 Development Notes

• All async endpoints tested using pytest-asyncio
• Follows PEP8 and FastAPI best practices
• Uses SQLAlchemy ORM + Pydantic models
• Consistent hashing and validation logic in utils.py
• Full reproducibility via Docker

---

✅ Definition of Done

• All tests pass (pytest)
• Docker Compose builds without errors
• API routes reachable at http://localhost:8000/docs
• Admin seed created successfully
• Model predictions functional
• User CRUD + Feedback endpoints verified

---

📜 License

© 2025 — Developed for the ANYONE AI – MLOps Assignment 3 Author: Jose Daniel Soto