davanstrien/uv-scripts-for-ai

uv-scripts-for-ai is a collection of self-contained Python scripts for data and machine learning tasks like OCR, image analysis, audio transcription, and dataset creation. Each script is built using a tool called UV, which lets a single Python file list its own dependencies right at the top, so anyone can run it without cloning the whole repository, setting up a virtual environment, or installing a requirements file first. The scripts are called recipes. Most recipes read data from the Hugging Face Hub, a place where people share datasets and models, and write their results back there too, so the output of one recipe can become the input for the next. This makes it possible to chain several recipes together into a pipeline, for example turning a folder of scanned documents into searchable text, then feeding that text into another script for further processing. You can run a recipe on your own computer if it has the hardware needed, often a graphics card, using a command line tool called uv. If you do not have that hardware, you can instead point Hugging Face Jobs at the script's web address and it will run on managed cloud hardware, billed by the second, with no subscription required. Either way, there is no separate installation step: the dependencies are baked into the script itself and installed automatically when it runs. The repository organizes its recipes by domain, including optical character recognition, computer vision, audio, embeddings, data processing, dataset creation, synthetic data generation, running open language models over a dataset, and entity extraction from text. A skill file is also included so that AI coding agents can discover, run, and adapt these recipes on their own. This project would suit someone who wants ready made data and machine learning tools without writing them from scratch, particularly people already working with Hugging Face datasets. It assumes some comfort with the command line but tries to remove the usual setup burden that comes with running someone else's Python code.