1
00:00:00,760 --> 00:00:04,590
This first lecture will be an
introduction to LangChain.

2
00:00:05,060 --> 00:00:09,150
You'll learn what it is, its use cases,
and its components.

3
00:00:09,760 --> 00:00:13,210
After this theoretical lecture, we'll
move on to coding.

4
00:00:14,420 --> 00:00:15,490
What is LangChain?

5
00:00:16,140 --> 00:00:18,390
LangChain is an open -source framework

6
00:00:18,400 --> 00:00:25,470
that allows developers working with AI to
combine large language models or LLMs

7
00:00:25,480 --> 00:00:30,650
like GPT -4 with external sources of
computation and data.

8
00:00:31,740 --> 00:00:37,610
The framework is currently offered as a
Python or a JavaScript package and is

9
00:00:37,620 --> 00:00:39,170
under heavy development.

10
00:00:40,520 --> 00:00:43,830
It is used to build complete apps using

11
00:00:43,840 --> 00:00:45,450
LLMs very quickly.

12
00:00:46,660 --> 00:00:50,810
Using LLMs is often insufficient for

13
00:00:50,820 --> 00:00:52,710
creating truly powerful apps.

14
00:00:53,800 --> 00:00:57,270
This is because LLMs are often limited in

15
00:00:57,280 --> 00:01:03,690
their ability to understand the context,
interact with the real world, or learn

16
00:01:03,700 --> 00:01:04,610
and adapt.

17
00:01:05,700 --> 00:01:09,310
To understand what needs LangChain fills,

18
00:01:09,960 --> 00:01:11,750
let's look at a practical example.

19
00:01:12,860 --> 00:01:16,890
We all know that LLMs, such as ChatGPT

20
00:01:16,900 --> 00:01:19,790
and GPT -4, have an impressive general knowledge.

21
00:01:20,280 --> 00:01:23,230
We can ask them about almost anything and

22
00:01:23,240 --> 00:01:24,370
get a pretty good answer.

23
00:01:24,940 --> 00:01:26,650
But what if you want to know something

24
00:01:26,660 --> 00:01:30,470
specific from your own data, your own documents?

25
00:01:31,120 --> 00:01:33,870
It could be a book, a PDF file, or a

26
00:01:33,880 --> 00:01:36,450
database with proprietary information.

27
00:01:37,600 --> 00:01:40,290
LangChain allows you to connect a large

28
00:01:40,300 --> 00:01:44,890
language model like GPT -4 to your own
sources of data.

29
00:01:46,380 --> 00:01:49,070
This is what data -aware means.

30
00:01:49,680 --> 00:01:51,810
And LangChain is data -aware.

31
00:01:52,900 --> 00:01:58,550
Of course, you can give the information
to the AI model as input into the prompt.

32
00:01:58,900 --> 00:02:00,630
But I'm talking about something else.

33
00:02:00,640 --> 00:02:02,970
I'm talking about referencing an entire

34
00:02:02,980 --> 00:02:05,970
database field with your own data.

35
00:02:07,580 --> 00:02:11,010
Using LangChain, you can make GPT -4 work

36
00:02:11,020 --> 00:02:16,370
with your own data, and it can also take
the action you'd like.

37
00:02:17,500 --> 00:02:24,270
For instance, search the web, fill out a
form, or send an email with some specific information.

38
00:02:25,480 --> 00:02:27,930
This is what agentic -aware means.

39
00:02:28,940 --> 00:02:33,330
So, LangChain is data -aware and agentic -aware.

40
00:02:34,440 --> 00:02:38,810
The example I've described to you works
like this.

41
00:02:39,700 --> 00:02:44,950
You take the document you want your
language model to reference and slice it

42
00:02:44,960 --> 00:02:46,830
up into smaller chunks.

43
00:02:47,780 --> 00:02:50,710
Store those chunks in a vector database

44
00:02:50,720 --> 00:02:55,630
as embeddings, which are numeric
representations of the text.

45
00:02:55,640 --> 00:03:02,730
In fact, these capabilities open up an
infinite number of practical use cases.

46
00:03:03,240 --> 00:03:04,890
Just imagine the possibilities.

47
00:03:05,440 --> 00:03:07,870
You can have a large language model that

48
00:03:07,880 --> 00:03:12,170
references the Constitution or the Civil
Code of your own country.

49
00:03:12,480 --> 00:03:17,770
You can ask the application anything
about the Constitution or the laws you

50
00:03:17,780 --> 00:03:18,710
are interested in.

51
00:03:19,040 --> 00:03:21,210
And the application will give you the

52
00:03:21,220 --> 00:03:21,870
correct answer.

53
00:03:22,300 --> 00:03:25,310
It's like having a specialized lawyer in

54
00:03:25,320 --> 00:03:28,950
front of you that can be asked anything
for free.

55
00:03:29,440 --> 00:03:32,210
This will open up a world of new opportunities.

56
00:03:33,180 --> 00:03:35,310
I believe we are going to see exponential

57
00:03:35,320 --> 00:03:38,990
progress in data analytics and data science.

58
00:03:39,680 --> 00:03:41,970
Our ability to connect large language

59
00:03:41,980 --> 00:03:47,590
models to advanced APIs is really going
to take things to the next level.

60
00:03:48,600 --> 00:03:55,050
LangChain implements abstractions like
chains and agents so that a developer

61
00:03:55,060 --> 00:03:57,990
doesn't have to know what happens under
the hood.

62
00:03:58,740 --> 00:04:05,390
Let's go over some specific examples of
applications that can be created using LangChain.

63
00:04:06,020 --> 00:04:10,870
Chatbots – question -answering systems or
summarization tools.

64
00:04:11,720 --> 00:04:17,570
LangChain can be used to create chatbots
that can interact with users in a natural way.

65
00:04:17,940 --> 00:04:22,670
For example, you could use LangChain to
create a chatbot that can answer customer

66
00:04:22,680 --> 00:04:28,630
questions or provide support based on the
internal documentation of the company.

67
00:04:29,460 --> 00:04:34,390
LangChain can be used to create question
-answering systems that can answer

68
00:04:34,400 --> 00:04:37,790
questions in a comprehensive and
informative way.

69
00:04:37,800 --> 00:04:42,370
For example, you could use LangChain to
create a question -answering system that

70
00:04:42,380 --> 00:04:48,050
can answer questions about a particular
product, service, or topic such as

71
00:04:48,060 --> 00:04:50,790
history, science, or current events.

72
00:04:51,780 --> 00:04:53,710
LangChain can be used to create

73
00:04:53,720 --> 00:04:59,250
summarization tools that can summarize
long pieces of text in a concise and

74
00:04:59,260 --> 00:05:00,190
informative way.

75
00:05:00,660 --> 00:05:03,090
For example, you could use LangChain to

76
00:05:03,100 --> 00:05:07,650
create a summarization tool that can
summarize documents, news articles,

77
00:05:07,980 --> 00:05:10,210
research papers, or even books.

78
00:05:11,220 --> 00:05:13,510
These are only a few of the applications

79
00:05:13,520 --> 00:05:16,170
that can be built using LangChain.

80
00:05:16,840 --> 00:05:19,410
When learning LangChain, you will

81
00:05:19,420 --> 00:05:22,050
actually learn about three main concepts.

82
00:05:23,220 --> 00:05:26,790
LangChain components, chains, and agents.

83
00:05:27,660 --> 00:05:29,490
Let's see the LangChain components.

84
00:05:30,130 --> 00:05:33,590
First off, we have the LLM wrappers that

85
00:05:33,600 --> 00:05:40,530
allow us to connect to and use large
language models like GPT -4 or those from

86
00:05:40,540 --> 00:05:41,690
the Hugging Face Hub.

87
00:05:42,700 --> 00:05:45,930
Next, there are the prompt templates that

88
00:05:45,940 --> 00:05:50,570
allow us to create dynamic prompts which
are the input to the LLMs.

89
00:05:51,480 --> 00:05:56,390
Indexes allow us to extract relevant
information for the LLMs.

90
00:05:57,140 --> 00:06:04,170
And memory is the concept of storing and
retrieving data in the process of a conversation.

91
00:06:04,880 --> 00:06:09,530
There are two main types of memory, short
-term and long -term.

92
00:06:10,000 --> 00:06:15,070
Short -term memory generally refers to
how to pass data in the context of a

93
00:06:15,080 --> 00:06:21,510
single conversation and long -term memory
deals with how to fetch and update

94
00:06:21,520 --> 00:06:23,930
information between conversations.

95
00:06:24,980 --> 00:06:27,430
Chains allow us to combine multiple

96
00:06:27,440 --> 00:06:34,230
components together to solve a specific
task and build an entire LLM application.

97
00:06:35,380 --> 00:06:42,950
And agents facilitate interaction between
the LLM and external APIs.

98
00:06:43,440 --> 00:06:50,370
They play a crucial role in decision
-making determining which actions the LLM

99
00:06:50,380 --> 00:06:51,690
should undertake.

100
00:06:52,420 --> 00:06:56,030
Agents are enabling tools for LLMs.

101
00:06:56,940 --> 00:07:03,050
This process involves taking an action,
observing the result, and then repeating

102
00:07:03,060 --> 00:07:05,150
the cycle until completion.

103
00:07:05,980 --> 00:07:08,630
Ok, we'll take a break and in the next

104
00:07:08,640 --> 00:07:13,030
video, we'll move on to coding and go
deeper into each LLM concept.