00:00:00
in this video we're going to build
00:00:01
something from absolute scratch using
00:00:03
crew AI this is going to be a real world
00:00:06
use case this is something that I need
00:00:08
to build I'll explain what it is and I
00:00:10
will show you step by step exactly how I
00:00:13
go about building it a lot of this stuff
00:00:15
I don't know exactly how I'm going to
00:00:16
build I'm going to be testing it I'm
00:00:18
going to be making mistakes and I'm
00:00:19
going to share it all with you we're
00:00:20
likely going to be using open AI models
00:00:22
including GPT
00:00:24
401 we're going to be using perplexity I
00:00:27
have a lot of ideas about how this
00:00:29
should go and so let me tell you what
00:00:30
we're going to be building today
00:00:32
something that me and my team have been
00:00:33
thinking about is building out an
00:00:35
educational portal basically all the
00:00:37
information that you need to become
00:00:40
proficient at AI everything from the
00:00:42
very Basics to very complex tutorials
00:00:45
and we want to automate a lot of it for
00:00:47
at least the first drafts so we're
00:00:49
looking to build out usually text based
00:00:50
educational content and tutorials
00:00:53
there's going to be images as well
00:00:54
stepbystep guides and I want a crew to
00:00:56
put that together for me so with that in
00:00:58
mind let's get started started so I'm
00:01:01
going to try to use cursor today I have
00:01:04
not really used cursor a lot in the past
00:01:06
I'm usually using VSS code but let's use
00:01:08
cursor and we're going to take this as
00:01:10
far as we can today we might make this a
00:01:12
part one out of two or three we'll see
00:01:14
so we'll take it as far as we can so the
00:01:16
first thing we need to do let's just
00:01:18
spin up a new cond environment because
00:01:20
python environment management is hard so
00:01:23
cond create - nedu python equals 3.12
00:01:27
proceed yes all right now we're just
00:01:29
going to act activate the new
00:01:30
environment cond to activate edu there
00:01:32
we go now let's install crew AI so pip
00:01:36
install crew AI all right so it's
00:01:38
installing crew AI next we're going to
00:01:40
install length trce so pip install
00:01:42
length tr- python D SDK okay so this is
00:01:45
all going into that cond environment
00:01:47
that we just created again always use
00:01:50
Python environment management whether
00:01:52
it's VM or cond or something else
00:01:55
because python environment management is
00:01:57
the bane of my existence all right so we
00:01:59
got both of those installed now
00:02:01
installing crew and getting it set up
00:02:03
really could not be easier so what we're
00:02:04
going to type is crew AI create crew and
00:02:07
we're going to call it edu the edu crew
00:02:10
and then hit enter and so what this is
00:02:11
going to do is create the skeleton of an
00:02:13
app for us it's going to put together
00:02:14
all the files we need and it just makes
00:02:16
things really easy and here we go
00:02:18
already so it downloaded everything we
00:02:20
need select the provider to set up so
00:02:21
we're actually going to use a number of
00:02:23
different providers today but let's
00:02:25
start with open AI okay so select a
00:02:27
model to use we'll use GP t40 mini we
00:02:31
want speed and we want low cost at least
00:02:34
for the start we'll experiment with
00:02:36
higher end more expensive models soon
00:02:38
okay now we need an API key so I'm on
00:02:41
open AI create a new key we're going to
00:02:43
call this edu and I'll create secret key
00:02:46
copy paste it in hit enter and there we
00:02:49
go so you could see it created all the
00:02:51
files that we need right away now we
00:02:53
need to open up this folder where all of
00:02:55
these things were created so let's do CD
00:02:57
edu and then aside from that we click
00:02:59
open a folder we're going to go to the
00:03:01
desktop select edu and there we go we
00:03:04
now have our crew up and running then I
00:03:06
just make sure the edu conduit
00:03:08
environment is active so there it is so
00:03:10
it should be working now here we can see
00:03:12
we have main.py crew. py the config
00:03:15
includes agents. yaml and tasks. yl now
00:03:19
out of the box it comes with two agents
00:03:22
and two tasks that it knows how to
00:03:24
execute so let's just take a look and
00:03:27
we're going to run it just to make sure
00:03:28
it works so we have the res Arch task
00:03:30
and the reporting task and basically
00:03:32
what it's going to be doing is research
00:03:34
on a topic and then giving us a report
00:03:37
in markdown on that topic we can see the
00:03:40
agents right here we have a researcher
00:03:42
agent and a reporting analyst agent
00:03:45
let's see what happens so before we run
00:03:47
it let's do crew aai install hit enter
00:03:50
okay so it's getting everything we need
00:03:52
installed we notice a new environment
00:03:54
has been created do you want to select
00:03:56
it for the workspace folder now I always
00:03:59
get confused at this question because I
00:04:01
was already in an environment I believe
00:04:03
it's asking me what environment the
00:04:05
actual code editor should be in so I'm
00:04:06
going to click yes we might have to fix
00:04:08
that later unfortunately and yeah if
00:04:10
anybody knows how to make python
00:04:12
environment management easier let me
00:04:13
know in the comments cuz still after
00:04:15
almost two and a half Decades of coding
00:04:18
this is very difficult for me today's
00:04:20
video is brought to you by mamut mammut
00:04:23
AI brings all of the best models
00:04:25
together in one place for one price
00:04:28
Claude llama G GPT 40 mraw Gemini Pro
00:04:33
and even
00:04:34
gpt1 and rather than having to pay for
00:04:37
each of these AI separately you pay $10
00:04:40
to mammut and they bring it all together
00:04:42
in one place plus they have image
00:04:45
generation mid Journey flux Pro Dolly
00:04:48
and stable diffusion again all for $10
00:04:51
models are frequently updated as soon as
00:04:54
they're released so be sure to check out
00:04:55
mamut for access to all the best models
00:04:58
for one low price m m. a that is m a m m
00:05:02
o u t h. AI thanks again to mamut and
00:05:07
then let's just test it out crew AI run
00:05:10
let's see if it works so running the
00:05:12
crew perfect so it is looking like it's
00:05:13
going to work okay so we actually got
00:05:15
some warnings invalid escape sequence
00:05:19
interesting but it doesn't look like
00:05:21
that really matters all that much okay
00:05:24
here we go we have an agent doing the
00:05:27
research really basic stuff so far we
00:05:29
just want to make sure it's working all
00:05:31
right and if we go to the report. MD we
00:05:34
can see the report was created
00:05:36
successfully Perfect all right again
00:05:38
very basic we're not actually hitting
00:05:40
the web at all we're not using any tools
00:05:42
you can add tools custom tools right
00:05:45
here custom tool.i you can add built-in
00:05:48
tools from Lang chain there's a lot that
00:05:50
we can do we're not quite there yet all
00:05:52
right now that we know it's working
00:05:54
let's make sure we can get Lang trce
00:05:55
installed correctly so I'm going to go
00:05:57
to Lang trace. I already have an account
00:06:00
let's sign in I'm going to create a
00:06:03
project this is going to be called edu
00:06:06
and as the project type I'm going to
00:06:08
select crew AI create project Okay click
00:06:11
setup project generate API key so I'm
00:06:14
going to select this initialization code
00:06:16
right here copy and we're going to put
00:06:19
that in
00:06:20
main.py at the very top right below the
00:06:24
import okay so it looks like L Trace
00:06:26
python SDK is not installed we did
00:06:30
install it of course python environment
00:06:32
issue so I'm going to try using the
00:06:34
built-in AI so AI fix and chat here we
00:06:37
go install the module okay look how slow
00:06:40
that's going all right we'll come back
00:06:42
to that in a moment so I suspect the
00:06:45
environment that is being used for the
00:06:47
actual editor is not the environment
00:06:48
that we installed everything on because
00:06:50
we have it installed so now let's see
00:06:52
what environment we're using so I'm
00:06:53
going to hit command shift p we're going
00:06:55
to look for interpreter python select
00:06:58
interpreter and we're going to to look
00:07:00
for our edu so there it is right there
00:07:02
so let's select that and perfect now
00:07:05
it's not underlined that was the issue
00:07:07
and it should be installed now so let's
00:07:10
run it again and let's see crew aai run
00:07:13
so I didn't save it in time so I'm going
00:07:15
to abort that and let's try to run it
00:07:17
one more time this time I actually saved
00:07:19
it all right yeah we're still getting
00:07:21
Ling Trace no module found issue I ran
00:07:25
into this exact same thing yesterday I
00:07:27
don't really understand but let's see so
00:07:30
it definitely looks like we're on the
00:07:32
correct environment at least for the
00:07:34
editor now why isn't it being found is
00:07:37
the question so I'm going to hit cond of
00:07:38
list I think that's it let's look for
00:07:41
Lang trace and there it is so it is in
00:07:44
the list so why isn't this working all
00:07:46
right I don't know what to do here so
00:07:49
let's see what happens if I just
00:07:51
actually instead of doing crew AI run
00:07:53
let's see what happens if I just run it
00:07:55
from main.py might not be the yeah okay
00:07:58
no module Maybe crewp is the right place
00:08:00
to start it
00:08:02
from nope all right after battling with
00:08:06
python environments for the last 25
00:08:09
minutes I think I finally got it working
00:08:12
yes there we go okay so what I had to do
00:08:15
is essentially get rid of cond wipe the
00:08:17
VM environment start from scratch and
00:08:20
then just make sure that the python
00:08:22
version matched make sure that the Lang
00:08:24
Trace SDK was installed properly the C
00:08:27
module was installed properly and now
00:08:29
now it works there we go so we have a
00:08:32
new report right there now let's get
00:08:35
back into the edu project and Lang trace
00:08:37
and there we go finally we got it
00:08:39
working okay all right so the next thing
00:08:42
I want to do is get perplexity installed
00:08:45
I don't really want to deal with web
00:08:47
scraping cuz that's always really hard
00:08:49
to do and so I'm going to try to use
00:08:51
perplexities API and just let it do the
00:08:54
research for me now I've not done this
00:08:56
before with crew so we'll see if it
00:08:57
works so here's perplexity let's
00:08:59
generate a new API key copy now we have
00:09:02
perplexity so the first thing I need to
00:09:04
do is set up a new llm and in the crew
00:09:07
aai docs it actually tells you how to do
00:09:09
it so we should be able to do it pretty
00:09:11
easily so we're going to copy this code
00:09:13
right here let's switch back to cursor
00:09:15
and we should be able to just paste it
00:09:17
in right here I'm going to move this
00:09:20
import I don't need to say this again so
00:09:23
I'm simply going to grab llm and add it
00:09:25
to the end there and we're going to be
00:09:27
defining a new llm perlex
00:09:29
I definitely want to use a different
00:09:31
model than this and then we also need
00:09:32
our API key here I don't need this cuz
00:09:35
we're going to Define it below so two
00:09:38
things let's figure out which model we
00:09:40
want to use and then let's also input
00:09:43
the API key there all right so API key
00:09:46
is right there and now let's figure out
00:09:48
what model we want to use so right here
00:09:50
learn more supported models now this is
00:09:52
very unhelpful when I click supported
00:09:54
models this is the page it takes me to
00:09:55
so not great but let's see if we can
00:09:58
find out Which models are available docs
00:10:02
supported models okay there we go so
00:10:05
interesting we are only going to be able
00:10:08
to use these models okay let's start
00:10:13
with this one llama 3.1 sonar large this
00:10:16
is kind of the perplexity based models
00:10:19
so I'm just going to copy it we'll see
00:10:20
switch back to cursor we're going to use
00:10:24
there we go llama 3.1 all right so let's
00:10:28
see if we can get this to work now now
00:10:29
the researcher we're going to add a new
00:10:32
parameter here llm equals and I'm
00:10:35
actually going to rename this to
00:10:37
perplexity llm and look at that that's
00:10:40
so nice cursor so I'm just going to hit
00:10:43
tab tab and it just fills out everything
00:10:45
for me that's so nice all right so let's
00:10:49
give it a try let's just see if that
00:10:50
works so I'm going to run it again so
00:10:52
now the researcher should have access to
00:10:54
the perplexity API and hopefully Lang
00:10:57
Trace also captures it all right right
00:10:59
so we got some errors let's see what
00:11:01
happened so I'm not sure what happened
00:11:04
there so I'm going to just copy all
00:11:07
these issues and ask it to tell me
00:11:09
what's going on add to chat what's the
00:11:11
error getting a 404 error which doesn't
00:11:14
sound right no it it is it's inputting
00:11:18
the right code here so either we didn't
00:11:20
set this up right or possibly it's not
00:11:22
an open AI compatible API endpoint but I
00:11:25
I think it is all right so what I'm
00:11:26
going to do just to make sure it works
00:11:27
is let's switch back to the previous
00:11:31
version of what it had here so we'll use
00:11:32
mraw 7B instruct okay I'll hit save
00:11:36
let's see if we can get this to work and
00:11:38
then we know we've narrowed down the
00:11:40
issue nope same thing ah I see I deleted
00:11:43
this part okay so add that back in hit
00:11:46
save and let's run it again running the
00:11:48
crew all right error again perplexity
00:11:51
exception 404 all right let's just make
00:11:53
sure that we can actually hit the API
00:11:55
successfully so I'm going to say write
00:11:56
me code to test this API endpoint to
00:11:58
make sure it's working all right so I'm
00:12:00
going to copy all of this let's create a
00:12:03
new file paste it in and I'll save it
00:12:07
test.py great let's run it all right
00:12:10
yeah so we're still getting a 404 error
00:12:12
code why is that it's pretty obvious
00:12:14
we're hitting the perplexity API
00:12:16
incorrectly let's check out the
00:12:17
documentation all right so I'm going to
00:12:19
copy the code and actually ask
00:12:20
perplexity what I'm doing wrong all
00:12:22
right there are several issues with it
00:12:24
here is the corrected code all right the
00:12:28
correct endpoint is that fine messages
00:12:32
expects an array okay let's just see if
00:12:34
that works so we're getting a new error
00:12:35
at least invalid model okay that's true
00:12:39
so let's change out the model name now
00:12:41
all right so let's use this model again
00:12:43
let's see if we can get this to work so
00:12:45
here's for the model boom let's hit play
00:12:47
and I think that's working this time
00:12:49
there we go perfect okay so now we know
00:12:52
it works so let's see how we can get
00:12:55
this to work back in our crew code so
00:12:58
I'm going to to instead just put the
00:13:00
model name the base URL let's leave the
00:13:03
same let's see if we have to add chat
00:13:05
completions to it so let's do crew AI
00:13:08
run okay yes perplexity exception 404 so
00:13:13
SL chat SL completions let's see if that
00:13:16
fixes it one more time let's run it nope
00:13:19
that did not work okay so I'm going to
00:13:20
set it back now slv1 still let's try
00:13:25
asking perplexity again so now I say I
00:13:27
I'm trying to get crew AI to use the
00:13:29
perplexity API and getting a 404 error
00:13:31
using this code so enter let's see if
00:13:33
that can give us the solution okay
00:13:36
here's the corrected code the base URL
00:13:38
should be without V1 okay let's try that
00:13:41
let's remove the V1 right there okay add
00:13:44
the correct endpoint ah I see okay let's
00:13:48
add the correct endpoint hopefully that
00:13:51
is a thing and make sure you're using
00:13:53
one of the supported models wow I got to
00:13:55
say perplexity is excellent as a coding
00:13:58
assistant all right let's give it a try
00:14:00
now let's see if that works and if not
00:14:03
we'll pipe it back in and see if it can
00:14:06
correct it again so crew AI run let's go
00:14:10
is it working I think it might be
00:14:12
working it's definitely doing something
00:14:15
yes yes it worked amazing all right good
00:14:19
progress and let's see what we were able
00:14:22
to get out of it this time are we going
00:14:24
to get better research let's find out
00:14:26
okay so putting together everything now
00:14:28
good let's the report okay so we have
00:14:30
report. MD here we go comprehensive
00:14:33
report on current and emerging large
00:14:35
language models GPT 40 with multimodal
00:14:38
capabilities Falcon 180b which is kind
00:14:40
of old llama 2 llama 3 dbrx so basically
00:14:44
gave me a list of models that's okay all
00:14:48
right let's check the traces let's just
00:14:50
make sure everything looks good there so
00:14:52
I'll go over to Lang Trace let's refresh
00:14:54
the page and here we go so we can see
00:14:57
that it is using interesting Lama 3.1
00:15:00
one yep okay so that's right the large
00:15:02
versions it's using GPT 40 mini let's
00:15:05
click into it we can see the traces here
00:15:08
and then we can also see the cost over
00:15:11
here look at that absolutely just cheap
00:15:15
great great great okay so let's
00:15:18
experiment with some models now let's
00:15:20
use the bigger version of the perplexity
00:15:22
model to see if we get better results so
00:15:25
we'll take the huge version right there
00:15:27
and let's use use that right there we're
00:15:30
still going to use GPT 40 mini to put
00:15:33
together the report but let's see let's
00:15:36
actually take a look at what we're
00:15:37
getting back from perplexity and let's
00:15:40
go back to the trace so let's click in
00:15:43
let's see where it's actually grabbing
00:15:46
the perplexity here's open AI where's
00:15:48
perplexity okay so one thing that's not
00:15:51
actually clear is if the perplexity API
00:15:54
is actually using live web data because
00:15:57
all of the information it gave me like
00:15:59
nitron I guess that's more recent but
00:16:03
like Bloom and Falcon 180b these are old
00:16:06
models so let me see let's see if we can
00:16:08
change it to make sure we're getting the
00:16:10
latest and actually I'm going to ask
00:16:12
perplexity if it uses the web through
00:16:14
the API so does the perplexity API
00:16:16
search the web like the regular
00:16:18
interface does so it doesn't support Pro
00:16:20
search that's okay it does use the same
00:16:23
search subsystem it currently only
00:16:25
supports the sonar models the API search
00:16:28
cap capabilities are more limited okay
00:16:30
so that's fine maybe search is available
00:16:33
through the chat GPT API all right so is
00:16:36
chat GPT search available through the
00:16:37
open AI API let's see no it is not okay
00:16:41
so I think we're going to have to use
00:16:45
something else to do web search so let's
00:16:47
ask what's the best web search tool for
00:16:52
crew AI all right serper Dev tool EXA
00:16:55
duck duck search or duck ducko search
00:16:58
should say let's use serper I think
00:17:01
that's fine so we need to install crew
00:17:04
AI tools so let's do that so pip install
00:17:07
crew AI brackets tools hit enter nope
00:17:10
that did not work what what am I doing
00:17:13
wrong here okay ah I got it wrong okay
00:17:16
so there's no quotes there let's try to
00:17:18
install it again boom there we go okay
00:17:21
so now that that's installed first I'm
00:17:24
just going to comment out this
00:17:25
perplexity llm let's comment this out
00:17:29
okay so now it's using GPT 40 mini still
00:17:31
but now we're going to give it the
00:17:32
ability to actually search the web okay
00:17:34
so it actually gives us a good example
00:17:36
here so let's go ahead and copy that all
00:17:39
right so here we're going to paste that
00:17:43
so from crew AI tools import and we
00:17:47
really only need the serer dev tool so
00:17:49
let's get rid of these other ones for
00:17:51
now all right so we will need a serer
00:17:53
API key okay so serer dodev let's do
00:17:57
sign up and here we are okay let's see a
00:18:01
python example we don't really need that
00:18:04
we just really need the API key so let's
00:18:07
grab one copy all right so let's set
00:18:11
this so let's put the key there so we
00:18:14
set it as an environment variable let's
00:18:17
go ahead and put this under where we
00:18:19
actually call the serer tools we don't
00:18:21
have OS installed or called I should say
00:18:25
uh AI fix and chat and we just need to
00:18:27
import it yeah okay so we'll import it
00:18:30
at the very top under these boom so that
00:18:34
should work and now we have to
00:18:36
instantiate the tools so I'll copy that
00:18:39
under here put it right there okay
00:18:42
hopefully this is the right order we'll
00:18:43
fix it if it's not and now for our
00:18:46
researcher we're going to give the
00:18:47
researcher a tool tool Search tool
00:18:50
perfect okay now let's see what happens
00:18:54
if we just run it like that cre AI run
00:18:57
okay yeah look at that it is returning
00:18:59
search results beautiful so now we get
00:19:02
up-to-date information still kind of
00:19:04
boring the output it's giving but that's
00:19:06
okay we just want to make sure this
00:19:07
stuff is working now all right it's done
00:19:09
let's look at the report comprehensive
00:19:11
report on current trends in ai llms
00:19:13
multimodal ai agentic ai open source
00:19:16
llms small language models ethics
00:19:18
enhanced contextual length Okay I think
00:19:21
this is actually looking better much
00:19:22
better actually okay next let's give it
00:19:24
a topic I'm actually going to want to
00:19:25
use all right so we're not going to use
00:19:26
perplexity sorry perplexity you're out
00:19:28
of there we don't need this llm right
00:19:31
there right now all right so first I'm
00:19:33
going to get rid of this API key cuz I
00:19:35
don't want it in my actual code I'm
00:19:36
going to put it in the environment
00:19:37
variable okay so help me put this inm I
00:19:40
know how to do this I'm just going
00:19:41
through the steps here I kind of like
00:19:43
using the AI to make it foolproof for me
00:19:45
okay so add the API key so apply okay so
00:19:49
let's make sure it's in there boom look
00:19:51
at that that's so cool okay let's make
00:19:54
sure M has it it doesn't why don't you
00:19:57
have it there except cep load. M all
00:20:00
right so we need to import OS there we
00:20:02
also need to do Fromm load. M okay that
00:20:07
should be good now but it's not in here
00:20:10
oh there we go look at that cursor's so
00:20:12
good all right so we put all the keys
00:20:15
here we need one more key now so let's
00:20:17
go to crew and let's do that here so we
00:20:21
actually have to run install python. M
00:20:23
so we'll do that in a moment I'm going
00:20:24
say put this in the m file okay so serer
00:20:30
API key apply beautiful okay so yep
00:20:34
accept it once again we are going to
00:20:36
need to import M and then we need to add
00:20:41
this let's apply apply the key to M file
00:20:44
okay one click it's done all right good
00:20:47
okay so now we've gotten rid of all our
00:20:49
API keys from our main files and of
00:20:51
course let's add it to get ignore and it
00:20:53
already is okay good now let's run it
00:20:55
one more time just to make sure
00:20:56
everything works yes good good good now
00:20:59
while that's running let's look at some
00:21:00
of the traces all right so the last run
00:21:03
it is using GPT 40 mini only we see some
00:21:07
tool usage really nice look at that all
00:21:09
right and then let's see the cost too
00:21:11
it's very inexpensive right now now of
00:21:14
course if we use one of the 01 models
00:21:15
it'll become much more expensive but we
00:21:17
might get a lot better results let's see
00:21:18
what happens all right so it worked now
00:21:21
let's look at the report one more time
00:21:23
cost effective models weaponization
00:21:26
concerns okay so it actually gave me
00:21:27
something completely different cuz it's
00:21:29
such a broad topic that I defined in
00:21:31
main.py this is the topic AI llms so
00:21:35
broad now let's do something much more
00:21:37
specific all right so for the topic I'm
00:21:39
going to say basics of how retrieval
00:21:41
augmented generation Works let's run it
00:21:44
I'm going to do this once and then I'm
00:21:45
going to switch the model out for 01
00:21:47
mini and see if we just get a better
00:21:49
response out of it okay so doing a bunch
00:21:51
of web search perfect review the context
00:21:53
you got and expand each topic into a
00:21:55
full section for a report make sure it's
00:21:57
detailed and of course we haven't even
00:21:59
changed the definitions of the agents so
00:22:02
we'll take a look at that too actually
00:22:04
before this video is done all right so
00:22:05
it finished let's look at the report
00:22:07
retrieval augmented generation
00:22:09
definition and purpose good two-phase
00:22:11
operation Good integration with llms
00:22:14
okay this is actually pretty good it's
00:22:16
not that expansive so let's just look at
00:22:19
the definitions agents first your season
00:22:22
researcher senior data researcher let's
00:22:24
just say it's a senior researcher cuz I
00:22:26
don't want it to just be a data
00:22:28
researcher
00:22:29
and then uncover Cutting Edge
00:22:31
developments no so right incredibly
00:22:34
compelling educational content on this
00:22:36
topic so actually I'm going to add and
00:22:38
comprehensive boy I think I'm a convert
00:22:41
to cursor it is awesome you're a season
00:22:44
researcher with a knack for uncovering
00:22:46
the latest developments for uncovering
00:22:48
so it's not really the latest
00:22:49
developments so what I'm going to say
00:22:51
instead is you're a seasoned researcher
00:22:55
with a knack for putting together the
00:22:58
most
00:22:59
relevant information for educational
00:23:02
content on topic okay known for your
00:23:05
ability to find the most relevant and
00:23:08
comprehensive information and present it
00:23:10
in a clear and concise manner okay and
00:23:12
then for the reporting analist I think
00:23:15
I'm going to rewrite it as I'm going to
00:23:17
leave the name the same I don't want to
00:23:18
change it so let's say educational
00:23:21
content creator actually I am going to
00:23:23
change it here okay educational content
00:23:26
creator so since we Chang the name there
00:23:27
let's go to crew
00:23:29
.p and right there let's change it to
00:23:33
that okay yes educational content
00:23:36
creator boom boom okay I think that
00:23:40
looks good so we have the researcher and
00:23:42
the content creator now let's make sure
00:23:45
everything under here looks good create
00:23:47
detailed reports so create detailed and
00:23:50
compelling educational content based on
00:23:53
topic topic and research findings you're
00:23:56
a meticulous analyst no uh you're a
00:23:59
educational content creator with a key
00:24:01
eye for detail you're known for your
00:24:02
ability to turn complex data and topics
00:24:06
into clear and concise educational
00:24:08
content making it easy for others to
00:24:09
understand and act on the information
00:24:10
you provide okay perfect let's go to
00:24:13
task let's see research task conduct
00:24:15
research make sure you find interesting
00:24:17
and relevant information for the given
00:24:19
year is 2024 okay expected output a
00:24:24
thorough research report on topic agent
00:24:28
okay reporting task review the context
00:24:31
you got and expand each topic into a
00:24:33
full section for a educational content
00:24:36
piece make sure the report is detailed
00:24:38
and no so I don't think it's a report
00:24:41
make sure the content is detailed yep
00:24:43
okay a fully fledged Report with the
00:24:47
main topics each with a full section of
00:24:50
information formatted as markdown great
00:24:53
reporting analyst that is not correct
00:24:56
gosh look at that look how it already
00:24:58
knows I should have switched that
00:25:00
wonderful all right let's give it a try
00:25:02
crew a run and then we're going to try
00:25:04
the 01 model okay so doing a bunch of
00:25:06
research McKenzie and Company Acorn Labs
00:25:08
we8 look at that wonderful getting
00:25:11
scraping in is easy it turns out okay
00:25:15
conclusion so got a bunch of information
00:25:17
put together a report great all right so
00:25:19
it's done let's go back to the actual
00:25:22
report and let's see what it did
00:25:24
understanding retrieval augmented
00:25:26
generation okay so it gives me a summary
00:25:28
at the top how it works great so
00:25:31
retrieval phase generation phase output
00:25:34
okay this is really good I want more
00:25:36
information though I want it more
00:25:38
comprehensive so let's see if we can
00:25:40
coax it into doing that and then I'm
00:25:41
going to switch it over to 01 all right
00:25:43
it says concise here so clear and
00:25:46
comprehensive I don't want concise all
00:25:49
right in a clear and
00:25:51
comprehensive manner let's make sure I
00:25:54
know I say comprehensive a few times so
00:25:56
let's just make sure doesn't have
00:25:59
concise anywhere else it doesn't okay
00:26:01
great now let's use the 01 model and see
00:26:04
if that works so for the researcher
00:26:07
that's fine we don't need a great model
00:26:09
for the actual research or maybe we do
00:26:12
let's do 01 mini across the board so llm
00:26:16
equals 01 mini it should just work like
00:26:18
this we have our researcher let's do it
00:26:21
here as well there we go okay and let's
00:26:25
run it again and then I'm going to check
00:26:27
Lang Trace to check the trace see what
00:26:29
it cost see how many tokens it used
00:26:31
let's see if it works all right so it is
00:26:35
working good good good agents final
00:26:38
answer so of course we're not going to
00:26:40
see The Chain of Thought cuz I think
00:26:42
that would break our interaction using
00:26:44
crew AI That's okay all right there it's
00:26:47
done good good good okay so now let's
00:26:50
check out the report wow look at this
00:26:53
very nice it gave us different types of
00:26:56
content this is a very very
00:26:59
comprehensive report benefits question
00:27:03
answering systems chat yeah this is by
00:27:06
far the best results so far
00:27:09
introduction amazing amazing yep okay
00:27:13
potential impact so this is fantastic I
00:27:16
think there's a couple things I want to
00:27:17
do but I'm going to save it for the next
00:27:18
video so one thing is let's try some
00:27:21
other models let's make sure we need the
00:27:23
01 model maybe we don't seems good
00:27:26
though I think I also want to implement
00:27:27
a a reviewer agent so not just output
00:27:31
something I want an agent to actually
00:27:34
review the content make sure it is
00:27:36
accurate make sure that it is explained
00:27:38
in a really simple way I also want to
00:27:41
try to get some images maybe we can
00:27:43
actually create some graphics on the Fly
00:27:45
that would be awesome so there's a
00:27:47
number of things that I want to add to
00:27:48
this so one more thing that I want to
00:27:50
show you before I wrap up this video
00:27:51
let's go to Lang Trace now we can see 01
00:27:54
mini right there so let's scroll across
00:27:57
this this one cost a lot more 58 cents
00:28:02
in output cost 4 cents in input cost so
00:28:06
you can see orders of magnitude more
00:28:09
expensive to use the 01 Mini model but
00:28:12
the report itself is also much better so
00:28:15
that's the tradeoff and I think what
00:28:16
we're going to have to do is just do a
00:28:18
bunch of testing different models
00:28:20
different price points different speeds
00:28:22
different costs and we'll see what
00:28:24
happens but for now I think that's it if
00:28:26
you enjoyed this video please consider
00:28:28
giving a like And subscribe and I'll see
00:28:30
you in the next one
