IBM and Google Launch Rival Toolkits for Document Intelligence

IBM's Docling is a Python toolkit that transforms PDFs, slide decks, and spreadsheets into structured formats like JSON or Markdown. It handles layout-aware tasks such as table extraction using models like TableFormer and DocLayNet. Released under the Linux Foundation’s LF AI & Data umbrella, Docling is designed to run locally on modest hardware, making it appealing to enterprise teams concerned with privacy and deployment flexibility.

Google’s LangExtract tackles the next layer up. It lets developers define structured schemas and use few-shot examples to extract information from text using Gemini models or local LLMs. Each extracted data point is grounded in the source text, offering traceability for use cases in legal, finance, and healthcare. LangExtract also includes interactive HTML visualizations for validating extraction accuracy.

These tools solve adjacent but critical problems in the AI pipeline. Docling turns varied document formats into clean, structured text that downstream models can process. LangExtract ensures that information pulled from that text is accurate, schema-compliant, and verifiable. Used together, they offer a modular foundation for scalable RAG workflows and document understanding systems.

By open-sourcing these tools, IBM and Google are reinforcing their commitment to developer ecosystems while expanding their AI portfolios. For enterprise AI teams, this marks a meaningful leap in operationalizing document intelligence with precision and transparency.