Hands-On AI: RAG using LlamaIndex

Learn how to enhance AI query capabilities and data accuracy through the application of LlamaIndex in retrieval-augmented generation processes.

Introduction

• Overcome the limitations of LLMs with RAG
• Limitations of LLMs
• Use cases for retrieval-augmented generation (RAG)

1. Getting Started

• Using GitHub Codespaces
• Setting up your environment
• Choosing an LLM and embeddings provider
• Setting up LLM accounts
• Choosing a vector database
• Setting up a Qdrant account
• Downloading our data

2. Fundamental Concepts in LlamaIndex

• How LlamaIndex is organized
• Using LLMs
• Loading data
• Indexing
• Storing and retrieving
• Querying
• Agents

3. Introduction to RAG

• Components of a RAG system
• Ingestion pipeline
• Query pipeline
• Prompt engineering for RAG
• Data preparation for RAG
• Putting it all together
• Drawbacks of Naive RAG

4. RAG Evaluation

• Introduction to RAG evaluation
• Evaluation metrics
• How to create an evaluation set

5. Advanced RAG: Pre-Retrieval and Indexing Techniques

• How we can improve on Naive RAG
• Optimizing chunk size
• Small to big retrieval
• Semantic chunking
• Metadata extraction
• Document summary index
• Query transformation

6. Advanced RAG: Post-Retrieval and Other Techniques

• Node post-processing
• Re-ranking
• FLARE
• Prompt compression
• Self-correcting

7. Modular RAG

• Hybrid retrieval
• Agentic RAG
• Ensemble retrieval
• Ensemble query engine

Conclusion

• LlamaIndex evaluation
• Comparative analysis of retrieval-augmented generation techniques