Go Beyond Basic RAG with Agentic Behavior
summary
Claims that extending RAG with agentic behavior lets an LLM decide when to rewrite queries, call tools, and run multi‑step retrieval, demonstrated by a Python pipeline built on Haystack that uses GPT‑4 mini as the decision model, Serp web search and a constrained Wikipedia tool for parallel retrieval, and Langfuse observability to show stepwise tool calls, documents saved to state, cost and latency metrics. The demo executes iterative and parallel tool calls, returns grounded answers, and leverages 160+ components and 70 integrations from the ecosystem.
AI-written summary · about
chapters (11)
- 0:00Introduction and agenda
- 2:05RAG refresher and benefits
- 6:40What is an AI agent?
- 8:47Agentic behavior explained
- 15:13Tool calling in agents
- 18:24Agentic RAG approach
- 19:29Haystack framework overview
- 25:16Demo setup and tools
- 29:34Building the agent pipeline
- 33:07Demo results and observability
- 35:37Q&A session
related talks
01h 30mOptimizing Retrieval for Agentic Systems
48mGoing Beyond Prompt Engineering
42mTo RAG or not to RAG
35mBuilding LLM-based Agents
27mOptimizing RAG Systems in Practice: Challenges and Proven Solutions
02h 47mFrom naive to advanced RAG: the complete guide
22mThe Power of RAG at Scale: A Year of Explosive Growth and Observab...
36mGraphRAG Explained 🔥 Building Smarter LLMs