Building Agentic RAG Systems

---

---

AI Summary

Summary of Video Transcript

Introduction

• Speaker: Erica Cardenas
• Role: Technology Partner Manager at we8
• Focus: Building end-to-end solutions and sharing them through we8 recipes and other mediums

Outline of Talk

1. Vanilla RAG vs. Agentic RAG
2. Agentic RAG and its benefits
3. Agent Ecosystem
4. Generative Feedback Loops

Vanilla RAG vs. Agentic RAG

• Vanilla RAG: Basic retrieve, augment, and generate pipeline using a vector index and language model.
• Agentic RAG: Adds layers to language models to make them agentic, including memory, planning, and tools.

Agentic RAG Benefits

• Understands user intent
• Calls tools in parallel
• Navigates databases
• Iteratively searches for information

Agent Ecosystem

• Building systems with large language models and function calling
• Agent Frameworks like llama index, DSP, Leta, linkchain, l graph, and a CI
• Adding observability with tools like Phoenix and Arise

Generative Feedback Loops (GFLs)

• API design for creating and storing new content using language models
• Applications: Cleaning data, chunking text documents, generating synthetic data

Resources

• Blog post and podcast by Erica Cardenas on agentic RAG
• we8 recipes with a subfolder for llm frameworks

Detailed Instructions and URLs

• No specific CLI commands, website URLs, or detailed instructions were provided in the transcript.