RecSysOps: End-to-End MLOps Recommendation System

Production-ready recommendation system demonstrating machine learning engineering, MLOps, containerization, CI/CD, experiment tracking, and cloud deployment.

Overview

RecSysOps is an end-to-end recommendation system built using the MovieLens 100K dataset. The project trains, evaluates, compares, and serves multiple recommendation algorithms through a FastAPI service deployed on AWS EC2 with automated testing and CI/CD.

Models Implemented

• Matrix Factorization
• Two-Tower Neural Recommendation Model
• Popularity Baseline

Features

• FastAPI inference service
• Docker containerization
• GitHub Actions CI/CD
• AWS EC2 deployment
• Experiment tracking
• Model comparison endpoint
• Automated testing
• Swagger API documentation
• Model artifact management

---

System Architecture

The diagram below illustrates the complete machine learning lifecycle from training to deployment.

Deployment Flow

MovieLens Dataset → Training Pipeline → Recommendation Models → Experiment Artifacts → FastAPI Service → Docker Container → AWS EC2 Deployment

GitHub Actions automatically runs tests and validates changes before deployment.

---

Screenshots

API Documentation

Interactive Swagger/OpenAPI documentation generated automatically by FastAPI.

---

Recommendation Endpoint

Example recommendation request and ranked movie recommendations returned by the API.

---

Model Comparison

Experiment tracking and model comparison across recommendation algorithms.

---

CI/CD Pipeline

Automated testing and deployment validation using GitHub Actions.

---

Model Evaluation

Models were evaluated using Precision@10 and Recall@10.

Model	Precision@10	Recall@10
Popularity	0.00477	0.04772
Matrix Factorization	0.00361	0.03606
Two-Tower	Training Loss = 0.378	N/A

Current best model according to Precision@10:

Popularity Baseline

---

API Endpoints

Health Check

GET /health

Response:

{
  "status": "ok"
}

Recommendations

POST /recommend

Request:

{
  "user_id": 1,
  "k": 10,
  "model_name": "matrix_factorization"
}

Metrics

GET /metrics

Returns training metrics, experiment artifacts, and model information.

Model Comparison

GET /compare-models

Compares available recommendation models and identifies the best-performing model based on evaluation metrics.

Model Information

GET /model-info

Displays available recommendation models and metadata.

---

Tech Stack

Machine Learning

• PyTorch
• NumPy
• Pandas
• Scikit-Learn

MLOps

• Docker
• GitHub Actions
• AWS EC2

Serving

• FastAPI
• Uvicorn

Testing

• Pytest

---

Project Highlights

• Trained and evaluated three recommendation approaches
• Implemented experiment tracking and artifact management
• Built production-style REST APIs using FastAPI
• Containerized the application with Docker
• Automated testing with GitHub Actions
• Deployed inference services to AWS EC2
• Added model comparison and monitoring endpoints
• Created an end-to-end MLOps workflow from training to deployment

---

Testing

Run locally:

pytest

---

Local Development

Clone the repository:

git clone https://github.com/farshad-haddadi/recsysops.git
cd recsysops

Build the Docker image:

docker build -t recsysops .

Run the application:

docker run -p 8000:8000 recsysops

Open:

http://localhost:8000/docs

---

Live Demo

Swagger UI

http://18.116.199.104:8000/docs

Health Check

http://18.116.199.104:8000/health

---

Example Results

Current deployed models:

• Matrix Factorization
• Two-Tower Neural Recommendation Model
• Popularity Baseline

The model comparison endpoint automatically evaluates available experiment artifacts and identifies the best-performing model.

---

Author

Farshad Haddadi

• GitHub: https://github.com/farshad-haddadi
• LinkedIn: https://www.linkedin.com/in/farshad-haddadi-b932a7346/