Docling's New Chart Data Extraction Feature Powered by IBM Granite

Peter W. J. Staar, a Principal Research Staff Member, Master Inventor, and Manager, announced a significant new Docling feature: state-of-the-art chart enrichment. This capability allows Docling to extract real numerical values directly from line, bar, and pie charts, converting them into clean, structured tabular data. This eliminates the need for screenshots or manual guessing, making data ready for analytics, Retrieval Augmented Generation (RAG), or agents. The feature is powered by the IBM Granite family of models and is completely open-source and permissively licensed.

To use it, users can run docling --enrich-chart-extraction <your-file>. A code example is available at https://lnkd.in/evXi7RvG, and the model is hosted on Hugging Face at https://lnkd.in/eqGUAdJc. The post encourages those interested in Document AI and multimodal understanding to try it and star the Docling project at https://lnkd.in/d4UT-6_2. Kudos were given to Rogerio Feris, Pengyuan Li, Abraham Daniels, David Cox, Sriram Raghavan, and Abdel Labbi for their contributions. The announcement hints that "charts are just the beginning," with more features coming soon.