Beginner’s Guide to

Vector Databases
AI by Hand ✍
Prof. Tom Yeh
Hosted by:

1
 Roadmap
+Vector
Database Retrieval

Word Dot Product

Embedding
Sentence
Embedding Transformer
Search Q/A
AI by Hand ✍ 2024 © Tom Yeh 2
Database
Beginner’s Guide to Vector Databases - AI by Hand ✍
 Fun fact

There are ___________________ millions dogs in the world!

AI by Hand ✍ 2024 © Tom Yeh 4

 How to create a table?
SQL:
__________ TABLE _________
( id __________,
name __________________,
size _________,
pop _________)

id name size pop

AI by Hand ✍ 2024 © Tom Yeh 5

 How to insert a record?
SQL:

_____________ INTO animals

_____________ (1, dog, 2, 900)

id name size pop

1 dog 2 900
AI by Hand ✍ 2024 © Tom Yeh 6
Vector Database
Beginner’s Guide to Vector Databases - AI by Hand ✍
 How to create a vector database?
SQL: CREATE TABLE animals
( id INT,
name VARCHAR(10),
size INT,
pop INT,
emb _________________ not null )

id name size pop

AI by Hand ✍ 2024 © Tom Yeh 8

 How to insert a record with a vector?
SQL:

INSERT INTO animals

VAUES (1, dog, 2, 900, ______________)

id name size pop emb

1 dog 2 900
AI by Hand ✍ 2024 © Tom Yeh 9
Retrieval
Beginner’s Guide to Vector Databases - AI by Hand ✍
 Which record is relevant to the query “cat”?

Query

cat id name size pop emb

1 dog 2 900 2 1 0
1 2 0
2 bat 1 10000 0 1 2

AI by Hand ✍ 2024 © Tom Yeh 11

 Draw distance vs similarity

distance similarity

AI by Hand ✍ 2024 © Tom Yeh 12

 Distance vs similarity on a scale of 1 to 5
asc or desc similarity

asc or desc distance

AI by Hand ✍ 2024 © Tom Yeh 13
 How to retrieve by similarity? (dot product)

________ name, emb<___>[__,__,__] AS score

FROM animals
________ BY ______ ASC | DESC ;l

AI by Hand ✍ 2024 © Tom Yeh 14

 How to retrieve by distance? (Euclidean)

SELECT name, emb<*>[1, 2, 0] AS score

FROM animals
ORDER BY score DESC;

AI by Hand ✍ 2024 © Tom Yeh 15

Dot Product
Beginner’s Guide to Vector Databases - AI by Hand ✍
 How to compute dot product?
Example:

1 2 3 dog 2 1 0
* * * * * *
2 2 0 cat 1 2 0
= = = ∑ = = = ∑
2 4 0 6
Result Result

AI by Hand ✍ 2024 © Tom Yeh 17

 How to compute dot product using matrix
multiplication?
Example:
dog
1 2
2 1
3 0

2 2 0 6 cat 1 2 0

AI by Hand ✍ 2024 © Tom Yeh 18

 How to compute dot product with multiple
vectors?
Example:
dog bat
1 1 2 0
2 1 1 1
3 1 0 2

2 2 0 6 4 cat 1 2 0 4

AI by Hand ✍ 2024 © Tom Yeh 19

Word Embedding
Beginner’s Guide to Vector Databases - AI by Hand ✍
 Where are dog, cat and bat in the “name”
space?

AI by Hand ✍ 2024 © Tom Yeh 21

 Where are dog, cat and bat in the “name”
space?

AI by Hand ✍ 2024 © Tom Yeh 22

 Which embedding is better?

Embedding 1 Embedding 2

dog cat bat dog cat bat

2 1 0 2 0 1
1 2 1 1 1 0
0 0 2 0 2 2

AI by Hand ✍ 2024 © Tom Yeh 23

 Which embedding is better?

dog cat bat dog cat bat dog cat bat

Desired Embedding 1 Embedding 2
2 1 0 2 1 0 2 0 1
dot
product 1 2 1 1 2 1 1 1 0
similarity 0 0 2 0 0 2 0 2 2

dog 2 1 0 H L dog 2 1 0 dog 2 1 0

cat 1 2 0 H L cat 1 2 0 cat 0 1 2
bat 0 1 2 L L bat 0 1 2 bat 1 0 2

AI by Hand ✍ 2024 © Tom Yeh 24

Sentence Embedding
Beginner’s Guide to Vector Databases - AI by Hand ✍
 How to embed sentences?

id comment user emb

1 How are you? John ?
2 Who are you? Mary ?

AI by Hand ✍ 2024 © Tom Yeh 26

 “How are you” à word embedding vectors
how are you

a an the how why who what are is am be was you we I they she he she me him her
0 -1 0 1 0 1 0 0 -1 1 0 0 0 3 1 0 -1 0 0 0 -1 0
2 0 2 0 0 0 -1 1 0 0 0 2 1 0 2 0 2 0 0 2 0 0
-1 0 -1 1 2 0 0 1 0 1 -1 0 0 -1 0 3 0 0 -1 0 2 -1
0 1 0 0 1 0 1 0 1 0 1 -2 0 0 0 1 0 1 0 1 0 1

AI by Hand ✍ 2024 © Tom Yeh 27

 Word vectors à Sentence vector
Method 1: Concatenate

how are you

1 0 0
0 1 1
1 1 0
0 0 0

AI by Hand ✍ 2024 © Tom Yeh 28

 Word vectors à Sentence vector
Method 2: Average

how are you

1 0 0
0 1 1
1 1 0
0 0 0

id comment user emb

1 How are you? John
2 Who are you? Mary
AI by Hand ✍ 2024 © Tom Yeh 29
 “Who are you” à word embedding vectors
who are you

a an the how why who what are is am be was you we I they she he she me him her
0 -1 0 1 0 1 0 0 -1 1 0 0 0 3 1 0 -1 0 0 0 -1 0
2 0 2 0 0 0 -1 1 0 0 0 2 1 0 2 0 2 0 0 2 0 0
-1 0 -1 1 2 0 0 1 0 1 -1 0 0 -1 0 3 0 0 -1 0 2 -1
0 1 0 0 1 0 1 0 1 0 1 -2 0 0 0 1 0 1 0 1 0 1

0 0
1 1
1 0
0 0
AI by Hand ✍ 2024 © Tom Yeh 30
 Word vectors à Sentence vector
Method 2: Average

who are you

1 0 0
0 1 1
0 1 0
0 0 0

id comment user emb

1 How are you? John [1/3, 2/3, 2/3, 0]
2 Who are you? Mary
AI by Hand ✍ 2024 © Tom Yeh 31
 How to query by SQL?

________ comment, emb<___>[__,__,__,__] AS score

FROM posts
ORDER BY ______ ASC | DESC ;

AI by Hand ✍ 2024 © Tom Yeh 32

 How to query using a high-level API?

query = Query(post_index)
._________(post)
._________(relevance_space.text, Param("________"))

app.query(query, _________ = "who are you?" )

Source: Superlinked.com

AI by Hand ✍ 2024 © Tom Yeh 33

Search
Beginner’s Guide to Vector Databases - AI by Hand ✍
 K-Nearest Neighbor, K=3, Dot-Product
Database
ID 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

emb

Query 9 -8 9 9 0 3 1 -6 0 11 3 13 -2 6 15 -9 7 6 -5 8

{ max | min }

AI by Hand ✍ 2024 © Tom Yeh 35

 K-Nearest Neighbor, K=3, L2
Database
ID 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

emb

Query 6 8 9 9 1 10 0 9 12 15 2 13 12 6 15 9 7 6 5 8

{ max | min }

AI by Hand ✍ 2024 © Tom Yeh 36

Transformer
Beginner’s Guide to Vector Databases - AI by Hand ✍
 How to use a Transformer to get a sentence embedding
vector?

Word Sentence
Embedding Embedding
Vectors Vector
1 0 0
0 1 1
1 1 0
0 0 0

AI by Hand ✍ 2024 © Tom Yeh 38

 How to combine across positions?
1
0
1
1 0 0
0 1 1
1 1 0
0 0 0

AI by Hand ✍ 2024 © Tom Yeh 39

 How to combine across positions?
1 0
0 1
1 1
1 0 0 1
0 1 1 1
1 1 0 1
0 0 0 0

AI by Hand ✍ 2024 © Tom Yeh 40

 How to combine across positions?
1 0 0
0 1 0
1 1 1
1 0 0 1 0
0 1 1 1 2
1 1 0 1 1
0 0 0 0 0

AI by Hand ✍ 2024 © Tom Yeh 41

 How to combine across features?

1 0 0
0 1 1
1 1 0
0 0 0
1 1 1

1 0 -1 0 1

AI by Hand ✍ 2024 © Tom Yeh 42

 How to combine across features?

1 0 0
0 1 1
1 1 0
0 0 0
1 1 1

1 0 -1 0 1 1 0 1
0 1 1 0 0
AI by Hand ✍ 2024 © Tom Yeh 43
 How to combine across positions and
features?
1 0 0
0 1 0
1 1 1
1 0 0 0 0
0 1 1 2 1
1 1 0 1 0
0 0 0 0 0
1 1 1
1 0 -1 0 1

AI by Hand ✍ 2024 © Tom Yeh 44

 How to use a Transformer to get a sentence embedding
vector?
1 0 0
Word
Embedding 0 1 0
Vectors 1 1 1
1 0 0 0 0
0 1 1 2 1
1 1 0 1 0 Sentence
Embedding
0 0 0 0 0 Vector
1 1 1
1 0 -1 0 1
0 1 1 0 0 2 3 1
0 0 0 1 1 1 1 1
AI by Hand ✍ 2024 © Tom Yeh 0 0 1 1 0 1 1 0 45
Q/A
Beginner’s Guide to Vector Databases - AI by Hand ✍
AI by Hand ✍ 2024 © Tom Yeh 47