Unlock Information from Tables, Images and Complex Docs
summary
Claims that Docling extracts structured data from tables, images and complex multi‑column layouts and produces a Docling document JSON with hierarchical metadata and bounding boxes. Demonstrations use the open‑source Python CLI/library, Hugging Face model weights including a small VLM (250M), and LangChain/LlamaIndex bridges to build a RAG pipeline that runs locally and runs on CPU. Reported table metrics sit near 90–95% table accuracy per benchmarks; limits include OCR language support, occasional table failures, and no built‑in access controls.
AI-written summary · about
chapters (16)
- 0:00Dockling and RAG systems
- 1:31Traditional PDF parsers limits
- 6:17Balancing accuracy and efficiency
- 12:38Dockling’s architecture
- 22:58Handling images and tables
- 31:28Table extraction limits
- 35:02HR documents use cases
- 40:13Exporting to JSON and Markdown
- 44:01RAG pipelines with vector DBs
- 50:29Document permissions
- 55:43Chunking strategies
- 1:00:07Structured output benefits
- 1:09:32Capturing headers and sections
- 1:15:27Production environments
- 1:22:46Visual grounding metadata
- 1:25:46Air-gapped environments
related talks
01h 19mLearn to Unlock Document Intelligence with Open-Source AI
01h 00mComplex Data Ingestion with Open Source AI
21mStructuring the Unstructured
48mIntegrating Docling With OpenSearch for Advanced RAG and Agentic Applications
38mBuild AI-Ready Search: Integrating Docling with OpenSearch for Advanced RA...
26mAdvanced Document Parsing for AI Workflows
22mStructuring the Unstructured : Empowering the AI Workflows
01h 13mDocument Q&A using LangChain