00:00:00
this is an Instagram AI agent that can
00:00:03
scrape anything for you first I'll show
00:00:05
you the demo and then explain how
00:00:07
everything works if we send a message to
00:00:10
this telegram bot and ask how many
00:00:13
followers does has and the Cen is my
00:00:17
Instagram username so we confirm it and
00:00:21
get that amount of followers not a lot I
00:00:24
know where do I from this is my
00:00:27
Instagram we'll confirm it and now it
00:00:31
will say that I'm from B essentially
00:00:33
it's scraping the personal information
00:00:36
about the location where Instagram was
00:00:39
registered can you tell me if I am
00:00:41
subscribed to Logan it's confirming
00:00:43
every action because we prompt it to do
00:00:46
that and when you see the back end
00:00:48
you'll understand how it works currently
00:00:50
it is scraping my following right and
00:00:54
analyzing if I am you know subscribed to
00:00:57
this person or not I am not I have just
00:01:00
to followings account that they follow
00:01:04
which is my personal account and my
00:01:05
wife's account so it should say no and
00:01:07
we can say then something like okay can
00:01:11
you send me a list of accounts that I
00:01:13
follow so it says no obviously the login
00:01:16
call is not in my subscription so should
00:01:19
send us Eugene Caden and Jenny
00:01:25
Caden okay so now it is sending us the
00:01:28
two accounts that I follow as well as
00:01:32
the suggested accounts again we prompted
00:01:35
to do that just for the reference so now
00:01:38
we can say can you analyze the most
00:01:39
viral post that I have should ask us if
00:01:42
we confirm again cin so each step is
00:01:46
confirming the username how many posts
00:01:48
we should analyze let's say 23 and it
00:01:52
should ask us so do we analyze it based
00:01:54
on the likes or amount of views uh let's
00:01:58
do View views so now it will scrape the
00:02:03
23 L post and based on the views
00:02:08
obviously so it will not scrape the
00:02:09
actual post only the reals because reals
00:02:12
have the views and provide us the top
00:02:17
um the top watched real obviously it's
00:02:20
you know not a lot so 146 but this is
00:02:24
correct so now we can say I want to get
00:02:28
a coffee um somewhere in
00:02:30
okay how many coffee shops you'd like me
00:02:33
to recommend so now it understood that
00:02:36
I'm looking for you know specific place
00:02:38
like coffee shops so let's say seven and
00:02:41
now it ask us if we want to you know
00:02:45
scrap for buy real post location or name
00:02:48
because again so maybe we can scrap for
00:02:51
the specific names that have the coffee
00:02:54
shop in the title but now we can say
00:02:56
just buy it will do the scraping based
00:03:00
on the information on the keywords like
00:03:03
coffee shop and Barcelona that are
00:03:05
available within a username bio and we
00:03:09
are doing that through a specific
00:03:11
advanced Google search and I'll explain
00:03:13
how it's done so this is how the agent
00:03:17
looks before going into the technical
00:03:19
setup I want to mention a few things you
00:03:22
probably watch a lot of the tutorials on
00:03:24
YouTube where people buil similar agents
00:03:27
on platforms like n8n and similar for
00:03:30
the most part they do not have any
00:03:33
monetary value and do not have any you
00:03:36
know real life applications and just
00:03:38
made for educational or demo purposes
00:03:41
which is fine but in this case this
00:03:44
agent will be sold to one of our clients
00:03:48
so I do believe that it's very important
00:03:51
for you to understand where and why AI
00:03:56
should be implemented right and not just
00:03:59
see some fancy builds and think that
00:04:02
okay so this is what I need before going
00:04:05
into the decision process that my agency
00:04:10
was going through to make a decision to
00:04:12
build an agent I want to mention another
00:04:14
thing I launching my own sauce it will
00:04:18
help you to build and sell production
00:04:22
ready AI agents and automations within
00:04:25
just minutes and not weeks for the
00:04:28
development and I'm not ch this is a
00:04:31
suite of AI solutions that my agency
00:04:36
built for our clients that we packaged
00:04:39
and modified in a way where anyone with
00:04:44
any technical
00:04:45
knowledge going through the course that
00:04:48
you'll have access to having 247
00:04:51
technical support and having all of
00:04:54
these things already built in can deploy
00:04:58
ready to publish Solutions within just
00:05:02
minutes and the solutions include things
00:05:04
like Omni Channel AI agent that can talk
00:05:07
through Instagram messenger WhatsApp SMS
00:05:11
email database reactivation with Voice
00:05:15
SMS personalized rvms Outreach which is
00:05:18
very unique customer engagement agent
00:05:21
you know inbound voice AI Prospect
00:05:23
enrichment and there are quite a few
00:05:26
different solutions there and again they
00:05:29
are to production they're not just
00:05:31
templates right so they're like complete
00:05:34
solutions that are already built on the
00:05:37
platform and I'm not sure this is a wh
00:05:39
labeled version modified white labeled
00:05:42
version of goha level that is powered
00:05:46
you know by the tools like voice low and
00:05:47
it and viy and so this is like the pl
00:05:52
the ultimate platform that you need to
00:05:55
get access to to be able you know to do
00:05:59
almost anything very fast with you know
00:06:02
unlimited amount of support okay so in
00:06:04
order to get the license right now you
00:06:07
would need to join the webinar so I have
00:06:10
not launched it yet the launch will be
00:06:12
actually on the webinar where I will
00:06:14
give all the participants and early bird
00:06:17
price so there's the first link in the
00:06:19
description once you register and you
00:06:21
attend the webinar so you'll have the
00:06:23
you know the early be price for the
00:06:26
license for the platform and I'll
00:06:28
obviously you know go through the demo
00:06:30
and we'll also answer all of your
00:06:33
questions now back to the actual build
00:06:36
on this channel I was showing you that
00:06:39
we were working with a business I'm
00:06:42
going to blow some information that is
00:06:45
primarily operating through Instagram
00:06:49
they receive a lot of the inquiries from
00:06:52
the users and they you know had quite a
00:06:55
few you know Setters that process all
00:06:57
the you know questions and try to set
00:07:00
appointments so we did multiple flows
00:07:03
with them so this are just a few so
00:07:06
these are just a few from the actual
00:07:08
pool and one of them so once people were
00:07:12
going through the actual builds
00:07:16
right we needed to know if the they are
00:07:21
qualified to continue talking to us or
00:07:25
not so are they a Potential Prospect
00:07:28
like qualified Prospect for the business
00:07:31
and the way we did that is through the
00:07:35
small automation here that is triggering
00:07:38
the web Hook Once the tag check reels is
00:07:41
applied so essentially when people
00:07:44
message the tag is automatically applied
00:07:48
and this web hook gets run it is a
00:07:52
simple make automation that is scraping
00:07:57
the Instagram profile of the user
00:08:00
through appify right and then actually
00:08:03
doing a few you know aggregation
00:08:06
modification things and through the llm
00:08:10
module checking how many views the
00:08:15
person has on average on the reals right
00:08:19
and if he is qualified so if he has more
00:08:22
than 5K views on average then we go here
00:08:26
and we're like adding the tag which is
00:08:28
you know qualified and if the user is
00:08:30
not qualified we're adding not qualified
00:08:33
tag and to just give you an idea this
00:08:37
scenario was ran
00:08:40
um
00:08:43
260,000 times okay and it's obviously
00:08:46
the runs like on the modules it's not
00:08:47
like the total runs but that's a
00:08:51
lot to so and you can ask Hey Eugene so
00:08:55
why are you not actually building this
00:08:58
thing on n8n
00:09:00
and this is one of the examples that I
00:09:01
was talking about like you need to
00:09:03
understand your use case obviously NN is
00:09:07
better for this scenario right it's
00:09:09
obviously better because you you not
00:09:11
going to you know pay $300 the they pay
00:09:14
right now for make and you know so so
00:09:19
and you can like essentially handle um
00:09:21
this things like um Json things like
00:09:24
Json processing through the coding block
00:09:27
and the reason we did that because the
00:09:29
company already is using make and they
00:09:32
wanted specifically us to build it
00:09:34
through make and teach them how to
00:09:36
manage it because we are not charging
00:09:38
for the management for this Solution
00:09:39
that's why it's very important for us
00:09:41
was to actually provide them what they
00:09:43
want okay so that was one of the actors
00:09:47
that we had here right so the second so
00:09:51
and we deployed it right so it's working
00:09:53
fine as you can see it was ran 72,000
00:09:56
times okay in two month so once we did
00:10:00
the project right um the client said Hey
00:10:03
Eugene I also would want to see the
00:10:08
location of the user right so besides
00:10:11
the fact that if this user is qualified
00:10:13
or not through the views they also to
00:10:16
wanted to qualify them based on the
00:10:19
location right and that's why in
00:10:22
telegram I was actually demoing you the
00:10:25
location so this is like obviously not
00:10:28
the location of the user so it's like of
00:10:30
the real location it's more like the
00:10:32
location where where the Instagram was
00:10:35
registered um and like how like this is
00:10:37
the location that the Instagram has so
00:10:40
it's not always accurate but from our
00:10:42
testing it's you know 85% that's you
00:10:44
know it's fine that's why we decided to
00:10:47
implement this and like to do that we
00:10:49
were using rapid API and here as you can
00:10:53
see like if we run Mr Beast and we turn
00:10:56
on the include about section here we
00:10:59
have you know in United States and we
00:11:01
have the DAT joint which is not know
00:11:03
data point that we're using but here in
00:11:04
the United States we do so the countries
00:11:07
that are not qualified that do not have
00:11:10
the buying power that the client wants
00:11:13
are you know assigning to a different
00:11:16
type of the agent so we build both of
00:11:18
the agents on voice flow and again so
00:11:22
you can go watch the video somewhere
00:11:24
here that where I was going through the
00:11:26
setup and so like we we decided to still
00:11:29
engage with this user but with a
00:11:31
different like freeb product to make
00:11:33
them sign up for the newsletter okay so
00:11:36
that was you know a sacent intergation
00:11:38
of the solution that we built then uh in
00:11:41
about a week um the owner actually came
00:11:45
up to me and said hey Eugene so um I
00:11:47
would want to analyze the some of our
00:11:51
you know students and some of our
00:11:53
competitors right and see what are they
00:11:56
posting and what are the results and
00:11:59
they're more interested in the ratio
00:12:02
between the views and comments right so
00:12:04
this is kind of what they're what what
00:12:07
what they're you know targeting as the
00:12:09
main metric and I said okay so we can
00:12:12
build you you know a simple you know air
00:12:14
table interface where you would you know
00:12:16
input your Instagram username and we'll
00:12:18
input you know how many posts you want
00:12:19
to analyze and here you know click Start
00:12:22
and if you go to the stats it will
00:12:24
populate the know 30 last post and then
00:12:28
you can export for them and they're like
00:12:30
using powerbi to uh to you know analyze
00:12:33
all the accounts because they have
00:12:35
multiple LS and just stuff like that so
00:12:37
and we built this and on and it and
00:12:40
because like in this case they didn't
00:12:42
want to go through know the setup
00:12:44
manually they don't need to understand
00:12:45
it because it's it's again so it's not
00:12:48
running consistently right it's more
00:12:50
like uh one a few times a week so as you
00:12:53
can see uh the scenario is running and
00:12:55
um the idea here is that it's like it's
00:12:57
looping through each and every post that
00:13:01
um we are trying to Target and here we
00:13:04
are you know actually get a result so
00:13:06
you can see Zero are the views because
00:13:09
um this account like goalie is primar
00:13:11
like posting the tax based content so
00:13:14
doesn't have the views and that's it
00:13:16
right so this kind of the solution
00:13:18
obviously this is in duplicate version
00:13:20
of the real thing and that's it and we
00:13:22
we obviously I advise them to you know
00:13:25
we can build the interface on air table
00:13:28
but they you know still use powerbi so
00:13:30
it's better for them to just you know
00:13:32
export it and import it there and we can
00:13:34
obviously connect this thing to powerbi
00:13:36
but yeah so that was like the third
00:13:40
iteration of the actual solution right
00:13:42
of the of the project right still with
00:13:45
the same client still around like the
00:13:47
same kind of business
00:13:49
processes and then um I decided you know
00:13:54
what so how about we will build you an
00:13:57
actual agent like a chart interface
00:14:00
where you will be able to chat with the
00:14:03
agent and retrieve anything that you
00:14:06
want and we will power it up with all
00:14:10
the possible scrapers obviously there
00:14:11
are a lot of them so if you go so on
00:14:13
rapid API it's kind of easier to see so
00:14:16
if you go here there are like a lot of
00:14:18
the end points that you can use right so
00:14:20
and we just got the you know the most
00:14:23
popular ones and for example like get
00:14:25
profile data has a lot of the data
00:14:29
references here right so it's not just
00:14:31
no buy or something so we build this
00:14:34
simple agent and again uh this is just
00:14:36
for the demo purposes uh for the channel
00:14:38
so when you're like sending a telegram
00:14:41
trigger right it's a pretty easy setup
00:14:44
so you just create a bot through bot
00:14:47
father on Instagram you're getting the
00:14:49
API key and you are putting it here at a
00:14:52
connection we are triggering this on the
00:14:55
message and then we are sending this to
00:14:58
this this message here through the
00:15:00
telegram so we are you know uh putting
00:15:03
here and then we're having a system
00:15:04
message this is how the system message
00:15:06
looks now for those who is kind of
00:15:10
familiar with NN
00:15:12
already might have a few questions so
00:15:15
the first is you know this is always how
00:15:18
we structure all of our prompts it's
00:15:20
obviously in the markdown it's obviously
00:15:23
divided by sections with the hashtags as
00:15:26
well as the separators we all always SE
00:15:29
like uh specify must never and always
00:15:33
type of the things that you know um in
00:15:37
the bolt and capital letter formatting
00:15:39
so it's so AI has a
00:15:42
better so that AI has an easier time
00:15:46
understanding the actual requirements as
00:15:48
you can see
00:15:49
here but we are you know structuring the
00:15:53
prompt a bit differently so you can see
00:15:56
here that for example when we were
00:15:59
saying like the modules are the actual
00:16:01
like the tools right and the tools are
00:16:03
connected here and all of these tools
00:16:06
are just HTTP requests right so to epy
00:16:10
and we are using primarily apify and not
00:16:13
rapid apis just because again the kind
00:16:17
they already have the apify account and
00:16:21
the only rapid API API we use is I
00:16:25
believe for the location which is the
00:16:27
actual this endpoint specific
00:16:28
specifically because the epy one taking
00:16:32
you know
00:16:33
forever and we are essentially like
00:16:36
asking um when processing you know the
00:16:38
data so we are asking so we are going
00:16:42
backwards through the from the
00:16:45
Json that we are sending to the tool
00:16:49
right so we're asking it to process the
00:16:52
data through the actual um through this
00:16:56
you know through this Json and then we
00:16:59
are instead of using the placehold
00:17:05
placeholder definitions where we can
00:17:07
essentially say what what is the
00:17:10
username so what is the variable we can
00:17:13
you know specify it in the
00:17:15
description we are saying hey in Json
00:17:18
body replace the username with the
00:17:21
username specified by the user right so
00:17:24
it's a bit controversial because usually
00:17:27
you put hey like get like this tool gets
00:17:30
the profile data and we're not doing
00:17:33
that CU
00:17:35
we must make sure that all the data is
00:17:40
accurate so it's better so if we're
00:17:43
balancing
00:17:45
between the like agent understanding
00:17:49
what tool to run and making sure that
00:17:52
the tool has the best info possible so
00:17:56
we're like choosing the actual like the
00:17:58
tool so because for example obviously if
00:18:00
you are and from all of our tests and we
00:18:03
were L like deployed and they already
00:18:05
like started to test it and it's like a
00:18:06
week ago so it's still we we don't have
00:18:09
like enough data um but it's working
00:18:12
great right so it has you know the 100%
00:18:16
accuracy like heat rate so you C you
00:18:19
could technically like say hey get you
00:18:22
know run this tool when you want the
00:18:24
Google search information but now
00:18:28
because this is
00:18:30
a advanced Google search it might find
00:18:35
some problems you know putting the
00:18:37
variables in the right places so we
00:18:39
decided to go with this route just keep
00:18:41
in mind why we're doing that and here
00:18:43
and by the way I forgot to mention that
00:18:45
this template will be available for free
00:18:47
right so if you want you know production
00:18:50
ready templates that you you know can
00:18:52
sell as I mentioned so you would need to
00:18:54
get the license for um my like platform
00:18:58
and you know in order to do that join
00:19:00
the webinar but a lot of the templates
00:19:03
that I share on this channel are free in
00:19:06
my resource Hub so there is you know a
00:19:07
link in the description if you go there
00:19:09
um you sign up you will get a you know
00:19:11
unique link to the newsletter oh sorry
00:19:14
to to the actual resource Hub and there
00:19:16
you'll find this template again it's
00:19:17
free so here we are in the query we are
00:19:22
using C so this is um so we're kind of
00:19:27
in the variable we are
00:19:30
putting what the user wants as well as
00:19:35
the actual location right as well as you
00:19:38
know with the pages we're using the same
00:19:40
thing
00:19:42
so this is one of the ways that you can
00:19:47
find information through the Instagram
00:19:50
right and not through the um through the
00:19:53
like General search okay and this is why
00:19:57
when we were ask asking it to Output the
00:20:01
info about the you know the coffee shops
00:20:04
so it was specifically looking for
00:20:06
coffee shops in Barcelona and not just
00:20:09
some you know random shops and we can
00:20:11
like say the same thing like um I'm W to
00:20:16
go to the gym okay sorry gy Jesus I
00:20:21
spelled it weirdly uh Jim you would like
00:20:23
to find the information about how going
00:20:25
to gy in a specific location yeah uh
00:20:28
yeah sure
00:20:29
I am in you know in like
00:20:33
Westland uh Westland
00:20:36
Michigan uh find you know like
00:20:40
five so it will you know start obviously
00:20:44
running this thing
00:20:45
okay this is that was right so the idea
00:20:49
of uh the agent and this is how it is
00:20:52
set up we have a know chbt for OMI and
00:20:57
we have a window buffer memory again so
00:21:00
it was kind of Behaving a bit
00:21:02
differently because like it was
00:21:04
clarifying What specifically we want to
00:21:08
an like what account and what's
00:21:10
confirming that and you can like fix it
00:21:12
in the prompt and this prompt obviously
00:21:14
will be inside the um inside this uh
00:21:18
sorry inside this automation as well as
00:21:21
attached separately if you want to use
00:21:23
okay guys so thanks for watching let me
00:21:25
know by the way if you want me to
00:21:28
because we are like planning to
00:21:29
integrate more solutions specifically
00:21:31
for this business also so if you want to
00:21:33
stay updated let me know in the comments
00:21:35
and have a good one peace
