MLOps.community

Does AgenticRAG Really Work?

56 snips
Dec 12, 2025
Guest
Satish Bhambri
In this engaging discussion, Satish Bhambri, a Senior Data Scientist at Walmart Labs, dives deep into the evolution of machine learning, exploring the transition from RNNs to transformers. He sheds light on the emergence of RAG systems and their ability to ground large language models, tackling issues like hallucinations. Satish explains the benefits of agentic RAG for creating specialized agents and the trade-offs between APIs and agents. He also shares insights on vector database selection and the importance of data freshness in recommendation systems.
Ask episode
AI Snips
Chapters
Books
Transcript
Episode notes
INSIGHT

RAG Grounds Models; Agentic RAG Specializes

  • RAG grounds LLMs by retrieving relevant documents so the model talks about real facts rather than inventing them.
  • Agentic RAG extends this by specializing agents for specific contexts and data to reduce hallucination.
ADVICE

Build Small, Focused Agents

  • Break monolithic RAG systems into small, focused agents to enforce separation of concerns and security layers.
  • Design agents per use case to improve scalability, governance, and cost optimization.
ADVICE

Create Dynamic Prompts From Retrieval

  • Generate dynamic prompts from retrieved documents rather than static prompts to reduce hallucination when LLMs create SQL or answers.
  • Use the retriever's context to construct the prompt passed to the LLM so outputs stay grounded.
Get the Snipd Podcast app to discover more snips from this episode
Get the app