Why Your RAG is Failing , And How Open Source Can Fix It
summary
Pipeline shows that messy PDFs — broken reading order, nested tables, images and lost metadata — cause RAG failures long before the LLM. A preprocessing stack using Docling + Heron + TableFormer converts PDFs/Word/images into JSON/markdown chunks that preserve layout, tables and OCR text; indexing with OpenSearch BM25+vector+RRF performs hybrid retrieval. A 140-page RHEL manual test with 50 queries raises context preservation from ~42% to 87% with Docling and to 94% with hybrid search, increases answer completeness ~43% and cuts hallucination ~21%.
AI-written summary · about
chapters (11)
- 0:00RAG failures and fixes
- 1:25RAG failures in production
- 4:39Dockling for document parsing
- 7:07Dockling’s structured chunking demo
- 10:40Baseline vs Dockling chunking
- 13:50Dockling for enterprise docs
- 17:45Multi-format docs with Dockling
- 18:49Hybrid search for retrieval
- 21:35Performance with Dockling
- 24:26Key takeaways and guide
- 25:22Q&A
related talks
36mDon't tell me RAG is easy
11mRAG Evaluation Is Broken! Here's Why (And How to Fix It)
19mImproving RAG Systems - Key Challenges and Solutions
48mIntegrating Docling With OpenSearch for Advanced RAG and Agentic Applications
01h 00mComplex Data Ingestion with Open Source AI
27mOptimizing RAG Systems in Practice: Challenges and Proven Solutions
21mBeyond RAG: Building Robust Knowledge Frameworks for Complex Multi-document Retrieval for Enterprise
37mBuilding advanced RAGs with Elasticsearch OpenSearch