The article outlines a process for creating a Docker-based machine learning inference service that includes automated CI/CD. Each push to the repository triggers a pipeline that scans for security vulnerabilities, builds the Docker image, tests it, and then pushes it to Docker Hub. The service operates as a Flask API that loads a Hugging Face model into memory and serves predictions without needing external API calls. The architecture ensures that everything runs locally within the container, which enhances reliability and performance. Significantly, the project includes automated security scanning with tools like Trivy and SonarCloud, and it employs multi-stage Docker builds, which help in creating smaller, more secure images. The CI/CD pipeline ensures that issues are identified before reaching production. The author shares that a critical vulnerability (CVE-2024–11392) was caught during the process, emphasizing the importance of security in machine learning deployments. Although the article provides a robust starter template, it also mentions gaps like the need for authentication, rate limiting, and monitoring, indicating that further development is necessary for production readiness. For practical use, the article provides a quick start guide that includes cloning the GitHub repo, testing locally, and configuring GitHub secrets. It highlights expected performance metrics, such as a cold start time of 40–60 seconds and warm inference response times between 100–300 milliseconds, with around 600MB of memory usage per container. The piece makes it clear that while this setup is beneficial for learning and development purposes, it may not be suitable for scenarios requiring lower latency or fully managed services.