Learn to Unlock Document Intelligence with Open-Source AI
summary
Docling is an open‑source Python library that actually converts messy PDFs docx excel powerpoint and HTML into a metadata‑rich Docling document format and exports markdown HTML or JSON, preserving reading order, merged table cells, images with captions and OCR. Lightweight models run locally (a 258M‑parameter VLM is available) so an eight‑page technical PDF converts in about 4.5s for eight pages during the demo. A hybrid chunker drops noisy fragment chunks (160→73) and produces bounded chunks for Milvus embedding, and a RAG pipeline links retrieved chunks to visual grounding boxes; Granite 3.8B runs on Replicate for generation.
AI-written summary · about
chapters (18)
- 0:00Dockling introduction
- 2:24Document structure importance
- 8:42Dockling pipelines
- 10:05Workshop labs setup
- 11:10Lab: docs to AI formats
- 13:37Document format and exports
- 16:27Tables and images handling
- 18:55OCR image processing
- 22:05Document structure visualization
- 23:08Image enrichment with VLMs
- 29:00RAG and chunking basics
- 37:49Dockling semantic chunking
- 40:38Chunking methods
- 45:11Chunk contextualization
- 56:21Full RAG pipeline
- 1:00:12Visual grounding for images
- 1:11:50Retrieval testing and responses
- 1:15:00Q&A
related talks
01h 19mLearn to Unlock Document Intelligence with Open-Source AI
01h 00mComplex Data Ingestion with Open Source AI
01h 07mMeet Docling: The “Pandas” for Document AI
31mDocling: Get Your Documents Ready for Gen AI
26mAdvanced Document Parsing for AI Workflows
29mConquering PDFs- document understanding beyond plain text
39mEnterprise-ready document processing for AI
33mUnlocking Insights From Multimodal PDFs Using OpenSearch and Vision-Language Models