Lecture 7

Presenter

Name: Spurthi Setty
Topic: RAG and Fine Tuning
Description: In this lecture, Spurthi Setty provides a deep dive into Retrieval-Augmented Generation (RAG) and fine-tuning as core building blocks for production-ready AI systems. The session explains how RAG connects large language models to external knowledge sources, enabling dynamic and up-to-date responses while reducing hallucinations. Spurthi differentiates between RAG, which focuses on improving factual grounding and retrieval, and fine-tuning, which shapes model behavior, tone, and structured outputs. The talk covers the full naïve RAG architecture—from document chunking and vectorization to similarity search and response generation—along with practical strategies for improving retrieval quality through better chunking, hybrid search, re-ranking, and vector database selection. The lecture also highlights evaluation frameworks such as Ragas to monitor performance and diagnose failure modes. Finally, Spurthi discusses the future of Agentic RAG, where autonomous systems iteratively retrieve, reason, and select tools, enabling more complex and multi-step decision-making workflows across domains such as finance, healthcare, and enterprise AI.

RAG is the default architecture for connecting LLMs to external knowledge, while fine-tuning is best used for behavior, tone, and structured output control.
Production-ready RAG systems depend heavily on high-quality chunking, embeddings, retrieval methods, and continuous evaluation.
Agentic RAG represents the next evolution, enabling autonomous reasoning loops, tool selection, and multi-step problem solving.