Aniket-k-13/AutoML-Studio

AutoML Studio is an ambitious no-code machine learning platform that brings many useful workflows into one browser interface. Users can upload datasets, explore data, preprocess features, compare models, train models, review results, export notebooks, and create prediction endpoints. The project also includes separate paths for tabular data, clustering, natural language processing, time series, and computer vision.

The range of features is impressive for a young project. The React interface is divided into clear pages for each stage of the workflow, while the FastAPI backend uses separate routers for uploads, training, evaluation, deployment, projects, Kaggle jobs, and exports. The repository includes Windows, macOS, Linux, Docker, and Hugging Face setup paths. It also has an architecture guide, changelog, release notes, contribution guide, issue templates, MIT license, and a live demo.

The README does a good job of explaining what users can build. It lists the available model families, evaluation metrics, preprocessing steps, and example use cases. The exported notebook feature is especially valuable for students because it can turn a no-code workflow into code they can study. The use of a non-root container user, streamed file uploads, configurable CORS origins, API keys for predictions, and request rate limiting are also positive choices.

The highest priority improvement is restoring automated tests. The repository currently has no test suite or CI workflow, and recent commits show that previous tests were removed. A platform with this many data paths needs tests for uploads, preprocessing, model selection, training, metrics, exports, deployments, Kaggle jobs, and frontend navigation. Small sample datasets could keep tests fast while still checking the complete workflow. CI should also run linting, dependency scans, backend tests, frontend tests, and a Docker build.

Several access controls need careful review before the public deployment is used with private data. Visitor identity is supplied through a request header, which a client can choose or copy. Some deployment operations do not confirm that the caller owns the related session. The undeploy route does not require an API key or visitor check, and the deployment status response can include the stored prediction API key. These routes should use server-issued sessions or real authentication, store only hashed API keys, and apply ownership checks to every read, update, deploy, export, and delete action.

Uploaded filenames should also be cleaned with a safe basename before they are joined to the upload directory. ZIP files need checks for path traversal, file count, extracted size, compression bombs, symbolic links, and unexpected file types. Image prediction uploads should have strict size and format limits. Error responses should avoid returning raw internal paths, command output, or provider details.

Model persistence deserves clear safety rules. Python pickle files can run code when loaded, so the application should only load files it created and should protect model directories from uploads or outside changes. Safer formats such as skops, ONNX, or framework-specific state files could be used where practical. Dataset retention, deletion, isolation, and privacy should also be documented, especially for the shared Hugging Face demo. Users need to know whether uploaded data is private, how long it remains, and who can access it.

The README says the entire pipeline runs in the browser, but training and storage actually happen on the backend. Updating that wording would give users a clearer picture of where their data goes. The project should also avoid describing the platform as industrial scale until authentication, testing, job isolation, and production monitoring are fully in place.

Overall, AutoML Studio is a creative and feature-rich educational project with a strong interface and broad machine learning coverage. Its next stage should focus less on adding models and more on tests, authentication, secure file handling, and honest deployment limits. Those improvements would make the platform much safer and more dependable without taking away from its excellent learning value.