mitmedialab/sherlock-project ? reverse-engineered prompt

Build me a small Python project that can look at a table and predict what kind of data each column contains, like name, address, city, or other semantic types.

I want it to work out of the box with a pre trained model, but also let me download the raw data, preprocess it, retrain the model, and run predictions on new tables. Please include a simple notebook or example flow that shows the full path from raw columns to features to prediction, so I can understand how to use it without digging through the code too much.

Make sure the project can be installed locally, loads the provided model files, and follows a clean scikit learn style interface for fitting and predicting. If anything depends on current library behavior, look up the latest docs online if you need to. Also include enough setup instructions so someone can run the examples and reproduce the training and test process from scratch.