1 | © Copyright 2024 Zilliz
1
1 | © Copyright 9/25/23 Zilliz
1 | © Copyright 9/25/23 Zilliz
Speaker
Jiang Chen
Ecosystem & AI Platform
jiang.chen@zilliz.com
@jiangc1010
2 | © Copyright 2024 Zilliz
2
Fantastic RAG Techniques
And Where to Find Them
Jiang Chen @ Zilliz
3 | © Copyright 2024 Zilliz
3
LLMs are great, but …
You still need to battle hallucination
with retriever, just like the Niffler
4 | © Copyright 2024 Zilliz
4
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 | © Copyright 2024 Zilliz
5
01 Review of RAG basics
CONTENTS
02 Advanced RAG techniques
RAG in action with Milvus Lite
03
6 | © Copyright 2024 Zilliz
6
01 Review of RAG basics
7 | © Copyright 2024 Zilliz
7
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- Knowledge of LLM is out-of-date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 | © Copyright 2024 Zilliz
8
9 | © Copyright 2024 Zilliz
9
02 Advanced RAG techniques
10 | © Copyright 2024 Zilliz
10
First thing first
Measure it before you attempts to improve it!
11 | © Copyright 2024 Zilliz
11
Indexing
Query Retrieval Prompt&
Generation
12 | © Copyright 2024 Zilliz
12
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 | © Copyright 2024 Zilliz
13
Query Enhancement
14 | © Copyright 2024 Zilliz
14
15 | © Copyright 2024 Zilliz
15
16 | © Copyright 2024 Zilliz
16
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 | © Copyright 2024 Zilliz
17
18 | © Copyright 2024 Zilliz
18
Indexing Enhancement
19 | © Copyright 2024 Zilliz
19
Good dishes come from good ingredients
• Data collection
• Data cleaning
• Parsing & Chunking
• DNN-native data?
20 | © Copyright 2024 Zilliz
20
21 | © Copyright 2024 Zilliz
21
Retriever Enhancement
22 | © Copyright 2024 Zilliz
22
23 | © Copyright 2024 Zilliz
23
24 | © Copyright 2024 Zilliz
24
25 | © Copyright 2024 Zilliz
25
Generator Enhancement
26 | © Copyright 2024 Zilliz
26
27 | © Copyright 2024 Zilliz
27
28 | © Copyright 2024 Zilliz
28
Agents!
29 | © Copyright 2024 Zilliz
29
30 | © Copyright 2024 Zilliz
30
31 | © Copyright 2024 Zilliz
31
32 | © Copyright 2024 Zilliz
32
03 RAG in action with Milvus Lite
33 | © Copyright 2024 Zilliz
33
34 | © Copyright 2024 Zilliz
34
Seamless integration with all popular AI toolkits
35 | © Copyright 2024 Zilliz
35
35 | © Copyright 9/25/23 Zilliz
35 | © 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 | © Copyright 2024 Zilliz
36
T H A N K Y O U

Advanced Retrieval Augmented Generation Techniques

  • 1.
    1 | ©Copyright 2024 Zilliz 1 1 | © Copyright 9/25/23 Zilliz 1 | © Copyright 9/25/23 Zilliz Speaker Jiang Chen Ecosystem & AI Platform jiang.chen@zilliz.com @jiangc1010
  • 2.
    2 | ©Copyright 2024 Zilliz 2 Fantastic RAG Techniques And Where to Find Them Jiang Chen @ Zilliz
  • 3.
    3 | ©Copyright 2024 Zilliz 3 LLMs are great, but … You still need to battle hallucination with retriever, just like the Niffler
  • 4.
    4 | ©Copyright 2024 Zilliz 4 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 Zilliz 5 01 Review of RAG basics CONTENTS 02 Advanced RAG techniques RAG in action with Milvus Lite 03
  • 6.
    6 | ©Copyright 2024 Zilliz 6 01 Review of RAG basics
  • 7.
    7 | ©Copyright 2024 Zilliz 7 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 Zilliz 8
  • 9.
    9 | ©Copyright 2024 Zilliz 9 02 Advanced RAG techniques
  • 10.
    10 | ©Copyright 2024 Zilliz 10 First thing first Measure it before you attempts to improve it!
  • 11.
    11 | ©Copyright 2024 Zilliz 11 Indexing Query Retrieval Prompt& Generation
  • 12.
    12 | ©Copyright 2024 Zilliz 12 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 Zilliz 13 Query Enhancement
  • 14.
    14 | ©Copyright 2024 Zilliz 14
  • 15.
    15 | ©Copyright 2024 Zilliz 15
  • 16.
    16 | ©Copyright 2024 Zilliz 16 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 Zilliz 17
  • 18.
    18 | ©Copyright 2024 Zilliz 18 Indexing Enhancement
  • 19.
    19 | ©Copyright 2024 Zilliz 19 Good dishes come from good ingredients • Data collection • Data cleaning • Parsing & Chunking • DNN-native data?
  • 20.
    20 | ©Copyright 2024 Zilliz 20
  • 21.
    21 | ©Copyright 2024 Zilliz 21 Retriever Enhancement
  • 22.
    22 | ©Copyright 2024 Zilliz 22
  • 23.
    23 | ©Copyright 2024 Zilliz 23
  • 24.
    24 | ©Copyright 2024 Zilliz 24
  • 25.
    25 | ©Copyright 2024 Zilliz 25 Generator Enhancement
  • 26.
    26 | ©Copyright 2024 Zilliz 26
  • 27.
    27 | ©Copyright 2024 Zilliz 27
  • 28.
    28 | ©Copyright 2024 Zilliz 28 Agents!
  • 29.
    29 | ©Copyright 2024 Zilliz 29
  • 30.
    30 | ©Copyright 2024 Zilliz 30
  • 31.
    31 | ©Copyright 2024 Zilliz 31
  • 32.
    32 | ©Copyright 2024 Zilliz 32 03 RAG in action with Milvus Lite
  • 33.
    33 | ©Copyright 2024 Zilliz 33
  • 34.
    34 | ©Copyright 2024 Zilliz 34 Seamless integration with all popular AI toolkits
  • 35.
    35 | ©Copyright 2024 Zilliz 35 35 | © Copyright 9/25/23 Zilliz 35 | © 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 Zilliz 36 T H A N K Y O U