Advanced Retrieval Augmented Generation Techniques

While achieving a basic Retrieval Augmented Generation (RAG) is relatively straightforward, attaining superior results requires tuning and optimizing various factors, such as a careful selection of embedding models. Additionally, applying advanced techniques, such as multi-stage retrieval with rerankers, is essential. A methodology for quality evaluation is also critical to success in crafting the best strategy for your specific use case. This talk will introduce the landscape of available optimization techniques and provide advice on best practices.

1. 1 | © Copyright 2024 Zilliz1 1| © Copyright 9/25/23 Zilliz1| © Copyright 9/25/23 Zilliz
Speaker
Jiang Chen
Ecosystem & AI Platform

[email protected]
@jiangc1010

2. 2 | © Copyright 2024 Zilliz2
Fantastic RAG Techniques
And Where to Find Them
Jiang Chen @ Zilliz

3. 3 | © Copyright 2024 Zilliz3
LLMs are great, but …




You still need to battle hallucination
with retriever, just like the Niffler

4. 4 | © Copyright 2024 Zilliz4
The evolution of AI made the semantic search of
unstructured data possible
Search by Probability
Statistical analyses of common
datasets established the foundation for
processing unstructured data, e.g. NLP,
and image classification
AI Model Breakthrough
The advancements in BERT, ViT, CBT
etc. have revolutionized semantic
analysis across unstructured data
Vectorization
Word2Vec, CNNs, Deep Speech pioneered
unstructured data embeddings, mapping the
words, images, videos into high-dimensional
vectors

5. 5 | © Copyright 2024 Zilliz5
01Review of RAG basics
CONTENTS
02Advanced RAG techniques
RAG in action with Milvus Lite03

6. 6 | © Copyright 2024 Zilliz6
01
Review of RAG basics

7. 7 | © Copyright 2024 Zilliz7
Why RAG?
RAG vs. LLM
-Knowledge of LLM is out-of-date
-LLM can not get your private knowledge
-Hallucinations
-Transparency and interpretability

RAG vs. Fine-tune
-Fine-tune is expensive
-Fine-tune spent much time
-RAG is pluggable

8. 8 | © Copyright 2024 Zilliz8

9. 9 | © Copyright 2024 Zilliz9
02
Advanced RAG techniques

10. 10 | © Copyright 2024 Zilliz10
First thing first
Measure it before you attempts to improve it!

11. 11 | © Copyright 2024 Zilliz11
Indexing
Query
Retrieval Prompt&
Generation

12. 12 | © Copyright 2024 Zilliz12
Types of RAG Enhancement Techniques
●Divide & Conquer
○Query Enhancement: better express or process the query intent.
○Indexing Enhancement: data cleanup, better parser and chunking
○Retriever Enhancement: more retrievers and hybrid search strategy
○Generator Enhancement: prompt engineering and more powerful LLM
●Thinking outside the box
○Agents? Other tools than retriever?

13. 13 | © Copyright 2024 Zilliz13
Query Enhancement

14. 14 | © Copyright 2024 Zilliz14

15. 15 | © Copyright 2024 Zilliz15

16. 16 | © Copyright 2024 Zilliz16
What are the differences in features
between Milvus and Zilliz Cloud?
Sub query1: What are the features of Milvus?
Sub query2: What are the features of Zilliz Cloud?

17. 17 | © Copyright 2024 Zilliz17

18. 18 | © Copyright 2024 Zilliz18
Indexing Enhancement

19. 19 | © Copyright 2024 Zilliz19
Good dishes come from good ingredients
•Data collection

•Data cleaning

•Parsing & Chunking

•DNN-native data?

20. 20 | © Copyright 2024 Zilliz20

21. 21 | © Copyright 2024 Zilliz21
Retriever Enhancement

22. 22 | © Copyright 2024 Zilliz22

23. 23 | © Copyright 2024 Zilliz23

24. 24 | © Copyright 2024 Zilliz24

25. 25 | © Copyright 2024 Zilliz25
Generator Enhancement

26. 26 | © Copyright 2024 Zilliz26

27. 27 | © Copyright 2024 Zilliz27

28. 28 | © Copyright 2024 Zilliz28
Agents!

29. 29 | © Copyright 2024 Zilliz29

30. 30 | © Copyright 2024 Zilliz30

31. 31 | © Copyright 2024 Zilliz31

32. 32 | © Copyright 2024 Zilliz32
03
RAG in action with Milvus Lite

33. 33 | © Copyright 2024 Zilliz33

34. 34 | © Copyright 2024 Zilliz34
Seamless integration with all popular AI toolkits

35. 35 | © Copyright 2024 Zilliz35 35| © Copyright 9/25/23 Zilliz35| © Copyright 9/25/23 Zilliz
Simplify and streamline
the conversion of
unstructured data into
state-of-the-art vector
embeddings, using
intuitive UI and Restful
APIs.
Pipelines
Easy. High-quality. Scalable.







Simplify the workflow
for developers, from
converting
unstructured data into
searchable vectors to
retrieving them from
vector databases
Deliver excellence in
every phase of vector
search pipeline
development and
deployment,
regardless of their
expertise
Ensure scalability for
managing large
datasets and
high-throughput
queries, maintaining
high performance with
min. customization or
infra changes
Zilliz Cloud Pipelines

36. 36 | © Copyright 2024 Zilliz36
T H A N K Y O U