Enterprise-ready document processing for AI
summary
Built an end-to-end pipeline that converts PDFs, images, audio and video into chunked embeddings, stores them in a vector DB and serves relevant passages to any LLM via RAG + vector DB. It runs on uploads → blob storage → Dockling processing → Vault tokenization for PII → embedding and indexing, supports diffing Dockling JSON across revisions, and keeps processing cheap so models don’t need retraining. Code and reproducible demo are available to clone.
AI-written summary · about
chapters (14)
- 0:00Introduction to speakers
- 0:52Hackathon robot project origins
- 2:58Why AI needs good docs
- 5:28Knowledge base challenges
- 7:54Limitations of current AI chat apps
- 11:23What is RAG and how it works
- 14:35Document processing with Dockling
- 16:22Reliability in AI pipelines
- 18:12High-level implementation overview
- 21:47Live demo setup
- 23:12File upload and processing demo
- 26:28Why not fine-tuning instead
- 28:34Three crucial lessons learned
- 30:41Q&A session begins
related talks
01h 00mComplex Data Ingestion with Open Source AI
26mAdvanced Document Parsing for AI Workflows
01h 25mLearn to Unlock Document Intelligence with Open-Source AI
01h 19mLearn to Unlock Document Intelligence with Open-Source AI
33mDeep research at scale
31mDocling: Get Your Documents Ready for Gen AI
31mConquering PDFs: document understanding beyond plain text
33mUnlocking Insights From Multimodal PDFs Using OpenSearch and Vision-Language Models