PyData Eindhoven 2024

The Levels of RAG 🦜
07-11, 11:15–11:45 (Europe/Amsterdam), Else (1.3)

This session's header image

LLM's can be supercharged using a technique called RAG, allowing us to overcome dealbreaker problems like hallucinations or no access to internal data. RAG is gaining more industry momentum and is becoming rapidly more mature both in the open-source world and at major Cloud vendors. But what can we expect from RAG? What is the current state of the tech in the industry? What use-cases work well and which are more challenging? Let's find out together!

Retrieval Augmented Generation (RAG) is a popular technique to combine retrieval methods like vector search together with Large Language Models (LLM's). This gives us several advantages like retrieving extra information based on a user search query: allowing us to quote and cite LLM-generated answers. Because the underlying techniques are very broadly applicable, many types of data can be used to build up a RAG system, like textual data, tables or even images.

In this talk, we will deep dive into this popular emerging technique. Together, we will learn about: what the current state of RAG is, what you can expect to work well and what is still very challenging.

Join us if you 🫵:
- Are interested in GenAI / LLM's and RAG
- Want to know more about the current state of RAG
- Would like to know when you can most successfully apply RAG

Contents of the talk 📌

  1. [2 min] Intro
  2. [4 min] Why RAG?
    1. The case for RAG
    2. The RAG advantage
    3. … so how-to RAG?
  3. [7 min] Level 0: Basic RAG
    1. Which ingredients make up a successful RAG system?
    2. Data ingestion
    3. Chunking
    4. Vector search
    5. Answer generation
  4. [6 min] Level 1: Hybrid search
    1. Combining multiple search methods with Reciprocal Rank Fusion
    2. TF-IDF
    3. BM-25
  5. [5 min] Level 2: Advanced data formats
    1. The landscape of data formats
    2. PDF parsing adventures
    3. Tables
  6. [3 min] Level 3: Multimodal
  7. [3 min] Summing things up
    1. The levels of RAG: from basic to advanced
    2. Concluding remarks
  8. [1 min] End

[30 minutes total]

❤️ Open Source Software

RAG and LLM’s are presented in a cloud-agnostic way. Many of the software libraries mentioned are open source. There is no agenda for representing any major cloud.

Prior Knowledge Expected –

The talk can be followed without RAG familiarity ✓. Interest in GenAI is enough ♡.
Jeroen Overschie

Jeroen is a Machine Learning Engineer at Xebia Data (formerly GoDataDriven), in The Netherlands. Jeroen has a background in Software Engineering and Data Science and helps companies take their Machine Learning solutions into production.
Besides his usual work, Jeroen has been active in the Open Source community. Jeroen published several PyPi modules, npm modules, and has contributed to several large open source projects (Hydra from Facebook and Emberfire from Google). Jeroen also authored two chrome extensions, which are published on the web store.