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
Building Production-Ready Search
Pipelines with Spark and Milvus
Jiang Chen @ Zilliz
3 | © Copyright 2024 Zilliz
3
AI revolutionizes how we use unstructured data
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
4 | © Copyright 2024 Zilliz
4
Retrieval-Augmented Generation
5 | © Copyright 2024 Zilliz
5
01 Why RAG is the New Search
CONTENTS
02 Search Pipelines with Spark and Milvus
The Future Landscape of RAG
03
6 | © Copyright 2024 Zilliz
6
Why RAG is the New Search
7 | © Copyright 2024 Zilliz
7
A Typical Search System
Picture Credit: https://web.eecs.umich.edu/~nham/EECS398F19/
8 | © Copyright 2024 Zilliz
8
Indexing
Query Retrieval Prompt&
Generation
Recap of RAG Architecture
9 | © Copyright 2024 Zilliz
9
Indexing
Query Retrieval Prompt&
Generation
Recap of RAG Architecture
Offline Indexing
10 | © Copyright 2024 Zilliz
10
Indexing
Query Retrieval Prompt&
Generation
Recap of RAG Architecture
Online Serving
11 | © Copyright 2024 Zilliz
11
How RAG Resembles Search
12 | © Copyright 2024 Zilliz
12
Search Pipelines with Spark and
Milvus
13 | © Copyright 2024 Zilliz
13
14 | © Copyright 2024 Zilliz
14
Easy to start with, can even run on edge devices!
15 | © Copyright 2024 Zilliz
15
Scale-up on Docker
16 | © Copyright 2024 Zilliz
16
Up to 100 billion vectors with K8s!
17 | © Copyright 2024 Zilliz
17
18 | © Copyright 2024 Zilliz
18
Using Milvus in LlamaIndex
Offline Indexing
Online Serving
19 | © Copyright 2024 Zilliz
19
Using Milvus in LangChain
Offline Indexing
Online Serving
20 | © Copyright 2024 Zilliz
20
Search Indexing with Spark and Milvus
21 | © Copyright 2024 Zilliz
21
Connect Spark Pipelines to Milvus through Connector
22 | © Copyright 2024 Zilliz
22
Connect Spark Pipelines to Milvus through Connector
https://milvus.io/docs/integrate_with_spark.md
23 | © Copyright 2024 Zilliz
23
The Future Landscape of RAG
24 | © Copyright 2024 Zilliz
24
Continuous Evaluation and Observability
Measure it before you attempts to improve it!
25 | © Copyright 2024 Zilliz
25
Multi-modal RAG
● Combining text and
image in the search
query
● Retrieving
multi-modal content
for generation
26 | © Copyright 2024 Zilliz
26
Good dishes come from good ingredients
• Data collection
• Data cleaning
• Transformation (Parsing, Chunking, Embedding, Labelling)
• Native data format of Neural Network?
27 | © Copyright 2024 Zilliz
27
Agents!
28 | © Copyright 2024 Zilliz
28
T H A N K Y O U
@jiangc1010

Building Production Ready Search Pipelines with Spark and Milvus

  • 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 Building Production-Ready Search Pipelines with Spark and Milvus Jiang Chen @ Zilliz
  • 3.
    3 | ©Copyright 2024 Zilliz 3 AI revolutionizes how we use unstructured data 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
  • 4.
    4 | ©Copyright 2024 Zilliz 4 Retrieval-Augmented Generation
  • 5.
    5 | ©Copyright 2024 Zilliz 5 01 Why RAG is the New Search CONTENTS 02 Search Pipelines with Spark and Milvus The Future Landscape of RAG 03
  • 6.
    6 | ©Copyright 2024 Zilliz 6 Why RAG is the New Search
  • 7.
    7 | ©Copyright 2024 Zilliz 7 A Typical Search System Picture Credit: https://web.eecs.umich.edu/~nham/EECS398F19/
  • 8.
    8 | ©Copyright 2024 Zilliz 8 Indexing Query Retrieval Prompt& Generation Recap of RAG Architecture
  • 9.
    9 | ©Copyright 2024 Zilliz 9 Indexing Query Retrieval Prompt& Generation Recap of RAG Architecture Offline Indexing
  • 10.
    10 | ©Copyright 2024 Zilliz 10 Indexing Query Retrieval Prompt& Generation Recap of RAG Architecture Online Serving
  • 11.
    11 | ©Copyright 2024 Zilliz 11 How RAG Resembles Search
  • 12.
    12 | ©Copyright 2024 Zilliz 12 Search Pipelines with Spark and Milvus
  • 13.
    13 | ©Copyright 2024 Zilliz 13
  • 14.
    14 | ©Copyright 2024 Zilliz 14 Easy to start with, can even run on edge devices!
  • 15.
    15 | ©Copyright 2024 Zilliz 15 Scale-up on Docker
  • 16.
    16 | ©Copyright 2024 Zilliz 16 Up to 100 billion vectors with K8s!
  • 17.
    17 | ©Copyright 2024 Zilliz 17
  • 18.
    18 | ©Copyright 2024 Zilliz 18 Using Milvus in LlamaIndex Offline Indexing Online Serving
  • 19.
    19 | ©Copyright 2024 Zilliz 19 Using Milvus in LangChain Offline Indexing Online Serving
  • 20.
    20 | ©Copyright 2024 Zilliz 20 Search Indexing with Spark and Milvus
  • 21.
    21 | ©Copyright 2024 Zilliz 21 Connect Spark Pipelines to Milvus through Connector
  • 22.
    22 | ©Copyright 2024 Zilliz 22 Connect Spark Pipelines to Milvus through Connector https://milvus.io/docs/integrate_with_spark.md
  • 23.
    23 | ©Copyright 2024 Zilliz 23 The Future Landscape of RAG
  • 24.
    24 | ©Copyright 2024 Zilliz 24 Continuous Evaluation and Observability Measure it before you attempts to improve it!
  • 25.
    25 | ©Copyright 2024 Zilliz 25 Multi-modal RAG ● Combining text and image in the search query ● Retrieving multi-modal content for generation
  • 26.
    26 | ©Copyright 2024 Zilliz 26 Good dishes come from good ingredients • Data collection • Data cleaning • Transformation (Parsing, Chunking, Embedding, Labelling) • Native data format of Neural Network?
  • 27.
    27 | ©Copyright 2024 Zilliz 27 Agents!
  • 28.
    28 | ©Copyright 2024 Zilliz 28 T H A N K Y O U @jiangc1010