Spotlight

Renumics Spotlight is an open-source Python library that enables interactive exploration of unstructured datasets directly from your pandas or Hugging Face dataframe. It provides a visual interface to inspect, label, and curate data, making it particularly useful for machine learning workflows involving text, images, audio, video, time series, and 3D data. Spotlight helps users understand data distributions, detect outliers, and identify quality issues through similarity maps and multidimensional projections. Targeted at data scientists, ML engineers, and researchers, Spotlight simplifies the iterative process of dataset debugging and curation. By leveraging embeddings and nearest-neighbor algorithms, it surfaces clusters, anomalies, and duplicates without requiring deep domain expertise in visualization. The tool integrates seamlessly with existing Python environments and supports custom embeddings for domain-specific analysis. What sets Spotlight apart is its focus on data-centric AI: it enables users to curate high-quality training sets by interactively selecting, filtering, and labeling samples based on similarity or manual criteria. The library is lightweight, extensible, and works with popular data formats like CSV, Parquet, and Hugging Face Datasets. It prioritizes user privacy by running entirely on the client side, ensuring no data leaves the local environment. While Spotlight excels at data understanding and cleaning, it is not a full-featured ETL platform nor does it replace larger data catalogs. Its strength lies in providing rapid, visual data inspection for small-to-medium datasets, making it an invaluable tool for exploratory data analysis and dataset preparation in machine learning projects.