brovk2008/dataset_collector

Dataset_Collector is a desktop application that lets you search, preview, and download datasets from multiple sources through a single interface. Instead of visiting Kaggle, Hugging Face, arXiv, bioRxiv, Dataverse, and other repositories separately, you run one search and get results from all of them at once. It is available as a standalone Windows installer that requires no Python installation, or as a pip package for macOS and Linux. The search system combines keyword matching with a local AI model to find datasets that are semantically related to your query, not just ones that share the exact same words. For example, searching for a topic can surface related research papers and dialogue datasets that a keyword search would miss. The AI model downloads once on first startup and then works entirely offline, so no data is sent to any cloud service after that. The application tracks your own search history and click patterns to gradually improve the ranking of results you tend to find useful. It also shows a "People Also Downloaded" list based on what other users have downloaded alongside a given dataset. A health score between 0 and 100 summarizes how complete and well-documented each dataset is, factoring in metadata quality, download availability, update recency, and popularity. Results and embeddings are cached locally in SQLite, which keeps repeated searches fast. The README notes a 100x speed improvement in deduplication and a 20x improvement in co-download logging compared to earlier versions. The project is tested on Windows, macOS, and Linux with Python 3.10 through 3.12. A continuous integration pipeline runs linting, type checking, and unit tests on every push. The current release passes all 19 unit tests and is available under the MIT License.