humansignal/label-studio

Label Studio is an open-source tool for labeling and annotating data so it can be used to train machine learning (ML) models. "Data labeling" means going through raw material, audio clips, text, images, videos, time series, and tagging it so an algorithm has examples to learn from. For instance, you might draw a box around every car in a set of photos so a model can later learn what a car looks like. The problem it solves is that most ML projects need large amounts of labeled examples before any training can happen, and producing those labels by hand is tedious without a dedicated interface. Label Studio gives teams a web-based UI for working through datasets across many data types, applying labels, and exporting the result in formats that fit common model-training pipelines. It can be used to prepare a fresh dataset from scratch or to clean up and extend an existing one to make a model more accurate. You run Label Studio locally or in the cloud. The README shows installs via Docker, pip, poetry, and Anaconda, plus a Docker Compose stack that bundles it with Nginx as a web proxy and PostgreSQL as the database, and one-click deploys to Heroku, Microsoft Azure, and Google Cloud. There is also an optional MinIO setup for S3-style local storage. You would use Label Studio when you are building or improving an ML model and need to create or curate the dataset that feeds it, anything from computer vision and text annotation to audio and time-series tagging. It is primarily written in TypeScript and Python, and the full README is longer than what was provided.