The Hidden Jewels to Solve Confabulation Challenges in Generative AI’s RAG Promise
summary
Claims concrete fixes: DataStax AI Platform packages Langflow and vetted libs and reports 50 companies moved rag into production serving 15 million users daily for one customer. Indexing detail matters — replace single-thread HNSW with G Vector disk ANN for synchronous writes and up to 32 processes, use Colbert token-level embeddings (≈10x vectors) only for high‑precision technical text, and prefer content‑centric graph augmentation (Astra DB graph) over basic rag when linked documents must be retrieved and reranked.
AI-written summary · about
chapters (9)
- 0:00RAG hallucinations overview
- 3:41Basic RAG limitations
- 5:55Vector indexing challenges
- 12:11ColBERT for technical concepts
- 16:59Graph-based RAG enhancement
- 23:51Basic vs graph RAG demo
- 28:53Low-code RAG apps
- 29:15DataStax GenAI program
- 31:04Q&A
related talks
56mRAG: from dumb implementation to serious results
49mRAG: from dumb implementation to serious results
01h 35mTutorial: Build AI-Supercharged RAG Apps with a Vector Database
53mBuild RAG from Scratch
27mOptimizing RAG Systems in Practice: Challenges and Proven Solutions
39mRAGs to Riches: How Our Content Affects Retrieval Augmented Generation
20mProduction RAG apps made easier with Astra
48mCrafting your own RAG system: Leveraging 30+ LLMs for enhanced performance