The Future of AI Assistants
summary
Argues that naive RAG fails for complex enterprise data and shows concrete fixes: an LLM-native parser (Lama Parse) that extracts tables, charts and captions to cut hallucinations; hierarchical indexing linking multiple text/image nodes to a single source; distributed embedding and ingestion using Ray to scale from thousands to millions of PDFs; an event-driven orchestration layer (LlamaIndex Workflows) that composes constrained agent ingredients or full agents with tool calls; and a managed stack (Lama Cloud) used in production for multimodal report generation and customer support automation.
AI-written summary · about
chapters (6)
- 0:00Introduction and agenda
- 3:54Problems with naive RAG
- 7:03High-quality data and retrieval
- 17:59Agentic reasoning over inputs
- 25:27Agentic output generation
- 28:16Future of knowledge assistants
related talks
32mBuilding Multi-Agent RAG Systems
32mBuilding an Advanced Knowledge Assistant
40mPractical Lessons from Deploying AI Assistants Based on LLMs
25mBuilding Your own AI Assistant with JavaScript, LLMs, and Rag
35mAdvanced RAG Techniques with LlamaIndex
18mLLMs and GenAI: A Guide to Building Your Own RAG Model for Document Interaction
47mBuilding and Testing AI-Agent Powered LLM Applications: A Live Demonstration
02h 08mIntroduction to AI Agents