Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated:
1
00:00:00,005 --> 00:00:01,009
- [Sudha] I'm going to show you
2
00:00:01,009 --> 00:00:04,008
how to build an autonomous agent AI
3
00:00:04,008 --> 00:00:07,007
using a no-code agent AI platform.
4
00:00:07,007 --> 00:00:09,004
There are several tools you can find
5
00:00:09,004 --> 00:00:12,005
in the Ecosystem handout that I've shared.
6
00:00:12,005 --> 00:00:15,002
I have chosen one called Promptly.
7
00:00:15,002 --> 00:00:19,005
You can sign up for a free account on trypromptly.com.
8
00:00:19,005 --> 00:00:22,008
Promptly provides a platform to build chatbots and agents
9
00:00:22,008 --> 00:00:27,000
using multiple LLMs that are available in the market today.
10
00:00:27,000 --> 00:00:28,004
Do you remember what is the difference
11
00:00:28,004 --> 00:00:35,007
between an intelligent chatbot and an agent?
12
00:00:35,007 --> 00:00:39,003
Well, a chatbot is an app that is built
13
00:00:39,003 --> 00:00:44,000
on top of an LLM model, and its learning is static
14
00:00:44,000 --> 00:00:46,005
and tied to the model that you choose.
15
00:00:46,005 --> 00:00:49,008
An agent can perform a complex task
16
00:00:49,008 --> 00:00:53,003
by breaking down our ask into simpler tasks
17
00:00:53,003 --> 00:00:56,001
and can learn from external data sources,
18
00:00:56,001 --> 00:01:00,000
and can run tools and reflect on its own results.
19
00:01:00,000 --> 00:01:03,000
Let's get started to build an agent.
20
00:01:03,000 --> 00:01:07,007
The first component to build an agent is a LLM.
21
00:01:07,007 --> 00:01:11,009
So, here you can see multiple options of LLM,
22
00:01:11,009 --> 00:01:14,009
so you need API access and connection to these LLMs
23
00:01:14,009 --> 00:01:18,003
if you start building the agent AI from scratch.
24
00:01:18,003 --> 00:01:21,002
I'm just going to choose OpenAI.
25
00:01:21,002 --> 00:01:24,003
And let's see whether we can make it
26
00:01:24,003 --> 00:01:26,007
do some image generation.
27
00:01:26,007 --> 00:01:34,008
Okay, so I'm going to say draw a marketing funnel.
28
00:01:34,008 --> 00:01:38,009
It is going to use DALL-E 2 as its model.
29
00:01:38,009 --> 00:01:40,005
I don't want such a huge image,
30
00:01:40,005 --> 00:01:42,006
so I'm just going to use a small one.
31
00:01:42,006 --> 00:01:46,002
So, we got it to generate an image.
32
00:01:46,002 --> 00:01:49,008
So, let me show text to speech,
33
00:01:49,008 --> 00:02:02,003
and we can say hello customers.
34
00:02:02,003 --> 00:02:03,006
- [Agent] Hello customers.
35
00:02:03,006 --> 00:02:05,007
- [Sudha] Okay, so here basically
36
00:02:05,007 --> 00:02:08,003
we have selected an LLM, so that's the first step.
37
00:02:08,003 --> 00:02:12,000
Next I'm going to show you in an agent-building process
38
00:02:12,000 --> 00:02:14,007
we can select outside data sources.
39
00:02:14,007 --> 00:02:16,007
So, I'm going to add a data source.
40
00:02:16,007 --> 00:02:18,006
You can have a single data store,
41
00:02:18,006 --> 00:02:21,005
it can be a text file or a PDF.
42
00:02:21,005 --> 00:02:23,004
I'm going to choose a URL.
43
00:02:23,004 --> 00:02:28,002
And for the purpose of this demo I'm going to make it go
44
00:02:28,002 --> 00:02:34,007
and scrape my own site called sudhajamthe.com.
45
00:02:34,007 --> 00:02:36,007
Oh, I need a name for it.
46
00:02:36,007 --> 00:02:39,004
I'm just going to call it SJ.
47
00:02:39,004 --> 00:02:40,004
Here we go.
48
00:02:40,004 --> 00:02:44,004
While it runs and loads I can add another data source.
49
00:02:44,004 --> 00:02:46,002
I'm going to give another one.
50
00:02:46,002 --> 00:02:51,000
This is my student community called businessschoolofai.com.
51
00:02:51,000 --> 00:02:53,004
Again, I'm trying to show you sites
52
00:02:53,004 --> 00:02:57,001
from where it's going to scrape and get information.
53
00:02:57,001 --> 00:03:00,003
And the idea is we are going to make the agent, again,
54
00:03:00,003 --> 00:03:02,001
I've not given it a name.
55
00:03:02,001 --> 00:03:05,005
And I'm going to call it Business School of AI.
56
00:03:05,005 --> 00:03:09,002
So, the idea with this data store is
57
00:03:09,002 --> 00:03:13,006
that we add external data sources, so it's ready.
58
00:03:13,006 --> 00:03:15,008
Here this is still loading,
59
00:03:15,008 --> 00:03:18,001
and it's going to get it ready in a minute.
60
00:03:18,001 --> 00:03:21,007
So, the agent has an LLM that it'll access.
61
00:03:21,007 --> 00:03:24,000
It could access more than one LLM.
62
00:03:24,000 --> 00:03:26,008
And it can access external data sources,
63
00:03:26,008 --> 00:03:29,007
so this could be a site, like I'm showing,
64
00:03:29,007 --> 00:03:33,004
it could be a file that is proprietary to your company
65
00:03:33,004 --> 00:03:35,003
that you could add that you want the agent
66
00:03:35,003 --> 00:03:37,003
to make referrals to.
67
00:03:37,003 --> 00:03:40,003
This is a RAG system that we would build
68
00:03:40,003 --> 00:03:44,003
that would be the agent that's built on top of that, okay?
69
00:03:44,003 --> 00:03:47,001
So, let that happen.
70
00:03:47,001 --> 00:03:57,003
So, now let me go to the actual building of the agent.
71
00:03:57,003 --> 00:04:01,006
If you know to build from scratch you can build from here,
72
00:04:01,006 --> 00:04:06,000
and you can say here is an agent I want to build.
73
00:04:06,000 --> 00:04:08,003
And the agent is called ASR,
74
00:04:08,003 --> 00:04:11,003
I'm just giving it some random name.
75
00:04:11,003 --> 00:04:12,009
And you can drag and drop the file here.
76
00:04:12,009 --> 00:04:14,007
Remember we added some data store,
77
00:04:14,007 --> 00:04:16,008
we can add files here also.
78
00:04:16,008 --> 00:04:19,005
I've not given API access, it's going to throw an error.
79
00:04:19,005 --> 00:04:22,001
But if you know to code and you are comfortable
80
00:04:22,001 --> 00:04:25,005
with the API access, I would recommend going this route
81
00:04:25,005 --> 00:04:28,003
and choosing the LLM here
82
00:04:28,003 --> 00:04:30,008
and adding what the LLM is going to do,
83
00:04:30,008 --> 00:04:32,006
all kinds of things, right?
84
00:04:32,006 --> 00:04:34,004
But there's an easier option
85
00:04:34,004 --> 00:04:37,002
because I want to show you a no-code option,
86
00:04:37,002 --> 00:04:39,007
that is what we are going to see a demo here
87
00:04:39,007 --> 00:04:42,007
because Promptly has provided a lot of templates
88
00:04:42,007 --> 00:04:46,009
that we can use readily to build this using no code.
89
00:04:46,009 --> 00:04:49,008
So, I'm going to choose Marketing,
90
00:04:49,008 --> 00:04:54,004
and let's create a hypothetical company, a dentist office,
91
00:04:54,004 --> 00:04:56,000
a new dentist office,
92
00:04:56,000 --> 00:04:58,001
and they want to build a marketing funnel.
93
00:04:58,001 --> 00:05:05,001
So, I'm going to ask this agent saying craft, not craft,
94
00:05:05,001 --> 00:05:16,004
build me a marketing funnel for a new dentist office.
95
00:05:16,004 --> 00:05:18,005
And what does the agent do?
96
00:05:18,005 --> 00:05:22,009
It takes my request, breaks it down to multiple steps,
97
00:05:22,009 --> 00:05:26,003
and then it starts coming up with a answer
98
00:05:26,003 --> 00:05:27,004
on how to build that.
99
00:05:27,004 --> 00:05:30,007
So, now it is telling me what are the components
100
00:05:30,007 --> 00:05:37,002
of the marketing funnel, and what does each section contain?
101
00:05:37,002 --> 00:05:39,007
So, we'll wait for a moment, it should be done.
102
00:05:39,007 --> 00:05:41,001
Yes, it's ready.
103
00:05:41,001 --> 00:05:42,007
Now, what I would like to do is
104
00:05:42,007 --> 00:05:45,006
I would like to see this as an image,
105
00:05:45,006 --> 00:05:50,009
show this marketing funnel as an image.
106
00:05:50,009 --> 00:05:52,009
This is generating an image,
107
00:05:52,009 --> 00:05:58,008
but from what it actually listed as the activities for us.
108
00:05:58,008 --> 00:06:01,008
So, it's not like the marketing funnel that we asked
109
00:06:01,008 --> 00:06:05,005
the LLM to generate, which was a random marketing funnel.
110
00:06:05,005 --> 00:06:06,006
This is the real thing.
111
00:06:06,006 --> 00:06:08,009
So, it takes my input message,
112
00:06:08,009 --> 00:06:12,006
create a marketing funnel for a new dentist office,
113
00:06:12,006 --> 00:06:16,003
if you see, it has updated the input message.
114
00:06:16,003 --> 00:06:19,007
The prompt I started from is create a marketing funnel,
115
00:06:19,007 --> 00:06:22,004
now it has now expanded that to create
116
00:06:22,004 --> 00:06:25,003
a marketing funnel image for a new dentist office.
117
00:06:25,003 --> 00:06:29,001
Since this is a continuation from the previous request
118
00:06:29,001 --> 00:06:31,005
where it listed all the options
119
00:06:31,005 --> 00:06:34,001
of what needs to get into the funnel, the six stages,
120
00:06:34,001 --> 00:06:36,002
so it's generated its own prompt.
121
00:06:36,002 --> 00:06:39,006
Can you see the agent has generated its own prompt saying,
122
00:06:39,006 --> 00:06:42,006
"Create a marketing funnel image for a new dentist office.
123
00:06:42,006 --> 00:06:45,000
The funnel should have six stages," and it's list that.
124
00:06:45,000 --> 00:06:46,004
"Each stage should include
125
00:06:46,004 --> 00:06:48,002
relevant channels and strategies."
126
00:06:48,002 --> 00:06:52,000
All this is what it came up with from its different
127
00:06:52,000 --> 00:06:54,003
step-by-step steps that it came up with.
128
00:06:54,003 --> 00:06:58,005
Now, it's putting all that generated its own prompt,
129
00:06:58,005 --> 00:07:03,005
and it's created a beautiful marketing funnel for us.
130
00:07:03,005 --> 00:07:06,000
I love it, awesome.
10404
Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.