Machine learning at scale 🤖

RAG Systems

Wondering how to optimize Retrieval-Augmented Generation (RAG) systems? This guide will take you through advanced RAG techniques to build efficient systems capable of handling complex query-response workflows.

From state-of-the-art retrieval algorithms to effective post-retrieval processing, this guide provides real-world case studies and actionable strategies to master RAG systems.

Below is the table of contents, offering a preview of topics covered in this extensive guide.

Looking to access the whole guide? Explore the premium option.

Full PDF (coming soon) Interested in a live high-end course? Let me know!

Table of Contents

I. Introduction

  1. Intro to RAG
  2. Failure Modes for RAG Systems
  3. End-to-End System Design we are going to implement

II. Retrieval Algorithms

  1. Branch and Bound
  2. Locally Sensitive Hashing
  3. Graph Algorithms
  4. Clustering Techniques
  5. Sampling Algorithms

III. Compression Algorithms

  1. Quantization
  2. Sketching

IV. Vector Retrieval in the Wild

  1. DiskANN
  2. Filtered DiskANN
  3. DESSERT
  4. FAISS
  5. Annoy
  6. ScaNN

V. Indexing

  1. Chunking Strategies
  2. Contextual Retrieval
  3. Document Summary Index
  4. Index Structure Optimization
  5. Metadata and Reverse HyDE

VI. Pre-retrieval Techniques

  1. Query Routing
  2. Query Rewriting
  3. Query Expansion
  4. Multi-query RAG Fusion

VII. Retrieval Optimization

  1. Keyword only, embedding only or hybrid?
  2. Late interaction

VIII. Post-retrieval Optimization

  1. Reranking
  2. Summarization
  3. Fusion
  4. Iterative / Recursive Retrieval

IX. Across the stack techniques

  1. HyDE (hypothetical question and answers)
  2. Sentence Window Retrieval
  3. GraphRAG
  4. Multi modal RAG and ColPali

X. Evaluation

  1. Precision, Recall, MRR, NDCG
  2. LLM as a Judge
  3. Unit Testing for RAG

XI. Comparisons

  1. RAG vs Long context
  2. RAG vs finetune (or both?)

XII. Use cases from companies

  1. System design of the world smartest email
  2. Productionizing LLMs
  3. LLM as a judge in production
  4. Knowledge graph with LLMs