Up next in 10
Conversation Building Block: Dapr in Action - Ep. 13
310 views
Jun 20, 2025
In this episode, you will understand what Conversation Building Block is, what are some of its concrete component implementations and how you can use it in your own applications. ABSTRACT A new weekly series about creating robust distributed applications and microservices using Dapr (Distributed Application Runtime). Developers and software architects who are building distributed applications and/or microservices and they want to understand how Dapr can make their lives easier. I will use .NET and C#, but will try to use other languages as well (for example, Python). šŗ CSharp TV - Dev Streaming Destination http://csharp.tv š C# Corner - Community of Software and Data Developers https://www.c-sharpcorner.com #CSharpTV #csharpcorner #Dapr #LiveShow
View Video Transcript
0:08
[Music]
0:19
[Music]
0:32
hey friends this is your host Rodrigo and today we're going to see the
0:38
conversation building block and this is the last session of the
0:44
series it's been a wild ride since December
0:50
and you have to take into account that all the different sessions are recorded
0:55
and you can find those in the cp.com
1:01
website and say that you want to I don't know you
1:08
want to continue your journey with Dapper there's of course a lot of different videos and training courses
1:15
that you can find online and okay that's it let me share my
1:24
screen which is number two okay cool
1:33
awesome let me start the
1:39
presentation so the conversation building block allows you to interact with llm
1:46
providers right such as open AI cloud or
1:52
all the different models that you can use in hugging phase and so on okay so
1:58
this is a a way for you to call those llms basically that's it take into
2:04
account that this is an it's an in an alpha state so all the things that I can
2:11
showed you today maybe they'll be changed in the future
2:17
okay um so let's go ahead and see the diagram which is this one you can see
2:23
that as any other building block you have your side car the side car
2:30
is the one responsible for calling the component of course and the component is
2:37
the one that is configured for a specific uh large language model and
2:43
provider okay today I'm going to show you open
2:48
Ai and of course you can use the same uh the
2:55
same uh techniques and skills for any other model
3:00
okay so this is nice I like the conversation building block the way that
3:07
the DI grade and the dopper community are thinking about it since you
3:12
can have some features that are not available in other Frameworks for
3:19
instance prom caching that's available in other Frameworks as
3:24
well but with promt caching it allows you to reduce costs and lat
3:30
by caching the prompts that are regularly sent to the
3:35
llms this is when you have a specific prompt that is recurrently being sent to
3:43
the LM that prompt is going to be cashed okay so of course that's going to reduce
3:51
cost in latency that's nice so in the other hand we have the pii ofation this
3:58
is personable identifiable information in other words
4:04
sensible or private information that you want to office gate you want to hide
4:11
when you're sending that data to the LM and the other way around as well when
4:17
you're receiving information that has pii this particular building block is
4:24
going to alus skate that data which is very nice I think that
4:30
approach is um interesting and of course just like any
4:37
other building block in Dapper this uh integration with large language
4:44
models are being observed and everything is going to be uh traced by using the
4:51
observability features that Dapper has which is quite fantastic
5:00
right so let's go ahead and see what we can
5:06
do with this okay let me go to the Dapper in action
5:14
folder and then this is the 13 folder since we are today in this
5:22
13th uh session and of course I don't have anything I just created the folder
5:29
uh some minutes ago so let's go ahead and create a new web API by using
5:36
asp.net and let's name it conversation so net new web API
5:43
conversation and let's use controllers you know
5:48
me I don't like uh minimal web apis so regular controllers and the name
5:57
is going to be conversation and basically that's it okay so for this
6:06
building block there's a nugget package that you can reference for calling the API remember the side
6:15
car your side card that is running alongside your application is exposing
6:20
an API that you can call by using HTTP or
6:27
grpc and you're free to uh you know call that end
6:32
points uh in any way you like right you can for
6:39
instance if you as you are a CP and net developer you can create your HTTP
6:44
client and then client do you know post async or get or put or those kind of
6:53
things however Dapper has fantastic s case that we can use and
7:01
net is one of the platforms that are supported in the upper and for this
7:08
particular conversation building block there's a specific nuget package this is not your regular Dapper asp.net core
7:16
nuget package so the name let's change the folder to here the name
7:25
is uh doer. a unsurprisingly I
7:32
think okay so I'm using net ad package of course you can use uh visual Studios
7:39
uh nuget ad package reference but since I'm here in the
7:44
terminal I just want to show you how to add this nuget package doer. a version
7:52
1.15.2 at the time of this recording and we can go ahead and verify
7:58
conversation cspr and the package reference is there
8:03
there's there's nothing new right okay so now let's create a new
8:11
solution and Net Solution add the current
8:16
project and we're ready to rumble conversation
8:22
sln and I'm going to open my visual studio 2022 instance
8:29
since you know me I'm a visual studio
8:35
Fanboy of course I use Visual Studio code from time to time but I think I
8:40
truly believe Visual Studio is better and is
8:45
superior uh to visual studio code but of course we can disagree right
8:52
you can disagree with me uh you can even be using other languages or Frameworks
8:57
that's fine so just like you're expecting here you
9:04
have to add the Dapper client but that's not the regular daer
9:10
client actually the same is the name is different Builder Services uh doer
9:16
conversation client okay this is the one so with this line of code you can
9:24
inject the dopper conversation client in your controllers and and you can even add
9:30
some configuration here you can pass this uh you know this expression an
9:38
action in this daer conversation client builder for setting up some things and
9:44
I'm going to use the defaults I don't want to focus on those details right now
9:49
I just I just want to show you that this is the way for injecting the D conversation client and let's get rid of
9:56
this nonsense whether forecast controller and let's rename this
10:04
to um say chat controller or conversation
10:10
controller okay this is good now using the primary
10:16
Constructor syntax I'm going to inject daer client daer conversation client
10:24
okay this is the one Dapper conversation client and of course it has some methods
10:31
and members that we can use for invoking those
10:38
LMS which llms you may ask well of course we need
10:44
the component we need to create the JAMA file for
10:50
sure so let me go ahead and open the browser and then developing applications
10:58
building blocks and then um
11:04
conversation and conversation overview of course you can read more about this
11:10
about the features what kind of information is considered uh private or sensitive
11:18
information such as phone numbers email addresses and so on it makes sense right
11:25
and let's navigate to those components you can see that AWS Bedrock is there
11:33
anthropic for calling CLA or those fantastic LMS are is there anthropic
11:40
deep siek is there deep siek such a fantastic
11:46
and uh glorious llm very interesting hogging phase
11:53
Mistral and of course open Ai and that's actually the one that I want to use used
11:59
today open Ai and you can see those required
12:07
values you know those fields that you you can specify the required one is key
12:14
that's the only one that is actually truly required API key for open
12:20
a okay take into account that this is again this is an alpha uh version okay
12:28
because it doesn't say anything about an endpoint so there's no Asher open AI
12:36
option right now of course Dapper is open source you
12:41
can contribute and create a component even if you don't contribute you can
12:47
create your own components uh anyway today I'm going to use the regular openi API the one from
12:55
open.com not Asher openi that is located in the Asher AI
13:02
Foundry uh umbrella in Asher so the key I have my key I created
13:10
the key beforehand because I don't want to spend so much time in that so I have
13:16
the key and the model the default is going to be GPT for
13:23
Turbo um but there are other um I think better models that we can use such as 40
13:31
40 mini right it depends on your requirements but today I'm going to use
13:37
GPT for old Mini I think it has the perfect balance between cost and
13:43
performance and logic and reasoning so GPT 40 mini and cash TTL this is the
13:52
time to live value remember that I told you that those proms um that you sent to the Ln
13:59
are going to be cashed by the building block so this is the Cash Time To Live
14:06
um duration okay the default is 10
14:13
minutes okay okay okay cool so good uh I think I'm just going to copy
14:21
this and return here this is my
14:26
application and this is uh my phone let's create components
14:33
components and let me go ahead and open a visual
14:40
studio instance here let's name it conversation.
14:47
jaml let me close those um panels those um
14:53
popups yeah I know I know okay cool so I'm gonna paste this
15:00
um so API version kind metadata and spec those are expected right Dapper doio B1
15:09
alpha1 is the API version the kind has to be
15:15
component and this is the name remember I told you like 1,000 times this is the
15:21
technical name that you want to use that you have to use in your
15:27
application so let's name it conversation can be called I don't know
15:33
open AI or so on this is just the name that I'm going to use today and the type the technical
15:41
type is conversation. openai okay this is the type for calling
15:48
those models and the key I happen to have my
15:55
key um right here please allow me to copy this and I'm
16:02
going to copy the entire string okay this is my
16:08
key let me show you where you can get that open.com I'm in this other monitor so
16:17
bear with me this is going to be fantastic um so open ai.com and then uh
16:25
API platform login
16:33
and I'm here yes I'm
16:38
here so I just logged in and you can go ahead and see your API
16:46
keys right here and of course you have to have enough uh
16:52
balance okay I have $10 and API keys this is the one drer in
17:00
action that I created beforehand and this is the one I'm going
17:05
to use this is it okay so model it's going to be gp40
17:13
mini you can always will go ahead and search for
17:21
open.com models and in this
17:26
page you can read more about those different models that open AI has
17:35
okay so many at the time of this recording those
17:41
are the ones and finally Cash Time to live this is fine um so this is
17:48
it okay let's go back to the application since I want to use doer conversation
17:56
client of course I need to create an point so public casing task let's say I
18:02
action res I action result let's name it
18:08
Chad okay this going to be HTTP [Music]
18:14
post and what about creating a record chat
18:20
request similar to this um um suggestion from
18:27
co-pilot but instead let's name it message so I have my record this is the
18:33
chat request chat request request right it comes from the
18:38
body this is this is just a regular asp.net core endpoint nothing new here
18:44
so doer conversation client dot you can see that we have Converse asnc Converse
18:52
async okay so convers acing takes the component name or some
19:00
unsurprisingly right this is conversation and then you have to pass
19:06
this I read only list of dopper conversation input and
19:13
then so this is this is an I read only list because you want to send the entire
19:20
conversation history to the large language model if you want to do that right if you want to keep those messages
19:28
or if you have to you know specify the system prompt and then maybe you want to
19:35
use uh few shot learning technique by using the assistant role
19:42
and then your prompt itself make sense so that's why that's an I read
19:49
only list and not only a string and so let's
19:56
pass a new topper conver ation input I'm going to pass request message I'm going
20:04
to pass uh the human you can see that this is an
20:09
inum system tool human function interesting I think the the
20:16
team is thinking about function calling and of course
20:24
agents um so maybe next in the next version we're going to see
20:29
those uh features in the Dapper Dapper composition building
20:36
block so okay human and I think that's
20:42
it what else oh if you want to scrub the private
20:49
information as you can see this is false you can pass through
20:54
and uh that information will be alus skated
21:00
so let's return something and now wait for
21:05
this and remember this is my I read only I'm
21:11
I'm using this um syntax from C 13 and I think that's
21:18
it we have outputs which is an I read only list of topper conversation
21:26
result and this is doer conversation input I don't like the
21:33
name uh conversation input drer
21:39
conversation output so maybe that's one of the things that I could say to the
21:45
team um daer conversation input and then result for me doesn't make sense it
21:52
should be named Dapper conversation output anyway so res do outputs I think that is
22:00
the one that I want and we also have this conversation
22:06
ID which is quite interesting since you want to
22:11
sometimes uh store all the different messages in a single thread right like the assistance
22:19
API that open AI created uh but in this case I'm just going to return let's say
22:27
the outputs just just like this and I think we're ready to
22:32
go Okay Okay cool so I think we're ready to test this
22:42
I'm ready to go ahead and return to the terminal
22:50
and let's go here and let's use toop run just like any
22:58
other application that we created in this series tap around and then the application ID it's going to be
23:05
conversation app or something conation app and then
23:10
dopper HTTP Port I want to use 55,000 and then a port I want to
23:18
be uh using this um 550 Z you know me I
23:25
always use that port and components path or resources path you you need to pass
23:32
that okay remember that the component is
23:37
located in this um folder and you know what let me
23:42
change the folder to this um conversation uh
23:47
project because otherwise I have to tell Don wrw to use that folder but I don't remember
23:54
the syntax right now and so components
24:00
PAAD it's going to be components so netron and then URLs
24:09
HTTP and it's going to be
24:16
55 so let's go ahead and execute this
24:22
remember I'm using components Pad but that
24:35
okay so this is running right so let's I don't know
24:43
let's use Curl or you can use posan or I want to use this conversation. HTTP file
24:51
that most of the time we're ignoring but this is so helpful since you can send those
24:58
requests to the endpoints right here and you have that in Source
25:04
control uh so 5500 and conversation is the
25:12
endo and I want to use content type application Json
25:17
and uh message right
25:24
message message is going to be hi
25:30
or something right for instance let's send a helpful prompt such as what are
25:37
the largest countries in the
25:43
world okay so you can see that you don't need to be executing this project you don't
25:51
need to be debugging the application you can send the request right
25:56
away so method not allow of course because this is not a a get is
26:02
supposed sorry about that send request and
26:11
then the largest countries in the world by total area are Russia and Canada
26:18
China the United States Brazil and so on so this is very
26:24
helpful and you can see that opening ey is the one that is responding specifically um g u GPT 40
26:32
mini is the one that is responding okay very very
26:41
helpful and I wonder about thatp uh
26:48
II so let's do the following let's see what we can do
26:55
here so comerse Asing
27:00
it has dopper conversation component name inputs and then
27:08
options okay options new what we have here let's see
27:15
conversation options there's nothing in the static
27:21
world so new conversation options and I can pass okay I can pass a
27:29
conversation ID very useful if you want to do something like you know
27:35
threads um or I don't know chat and
27:41
then threads right and then thread ID and use that thread ID right here and
27:48
pass that thread ID to this conversation options object um so that's the idea
27:58
what else we have here conversation options I think that's the only thing that we have today remember this is an
28:06
alpha version of the component
28:13
and uh so conversating inputs so here in inputs
28:21
you can specify that scrub pii let's turn it
28:27
into to True represents an input for the doer conversational
28:33
API uh if true scrubs the data okay okay so I think we're ready to test this
28:41
let's see what happens doer on again and let's go back and create another
28:52
prompt and I don't know uh let's see uh
29:01
I don't know let's um return what's the distance between one
29:09
Microsoft way that's an address in Redmond Washington
29:16
and uh I don't know uh let's
29:21
see and the safe C Is that the name of the stadium
29:30
in
29:35
Seattle so let's see the distance between one micr
29:40
away uh so it's not it's not obus skating maybe I'm
29:46
doing something wrong right so I don't know let's try another
29:54
thing uh change this personal information to
30:00
uppercase uh my name just like this I want everything in
30:05
uppercase let's see what's going on and send a request and it's not off
30:14
skating yeah I think something is missing in my code I'm setting this
30:19
personal identifi information uh property as
30:26
true and if true SC the data that goes into the LM yeah maybe I'm doing and this is not
30:33
the the correct demo for scrub pii uh anyway we have that there and
30:41
it's uh very useful we can refer to the
30:47
documentation anyway I want to go back
30:52
and um keep this um prompt that I was sending beforehand
31:00
okay and I think that's it I think that's it
31:05
and you can see that the conversation building block is quite useful uh but of course it's an an in an
31:14
alpha State uh if you want to build something more robust
31:21
or I don't know production ready maybe you can use this don't don't
31:28
get me wrong I think it's it's it's fantastic but maybe there are other
31:34
Frameworks more capable with more options if that's your uh
31:40
goal uh and of course we need to keep on
31:45
looking for those changes in the dopper conversation building block and let's
31:50
wait for the team and the community to uh evolve and continue uh polishing
31:57
this component which is fantastic I'm very excited to have this in Dapper for
32:05
um some um some things uh some practical
32:10
things um and but of course for from a
32:16
Enterprise point of view maybe I would use something else such as semantic kernel or other framework anyway this is
32:25
great this is fantastic great work in the Doer conversation building blog
32:30
and you know what this is the last session of the
32:40
series I'm so so happy that we made it through the end and if you made it to
32:46
the end as well congrats um I'm sure you know more about
32:53
Dapper than when you started back in December and some Shameless plug I have some
33:03
courses about Dapper in LinkedIn learning uh that you can
33:09
use and I'll be continue creating Dapper
33:15
content um and yeah yeah yeah um just uh
33:22
T take that into account and I think that's it I think that's it
33:29
and this is it friends this is it and thank you very much for the
33:35
c.com team and the entire crew for having me 13 sessions today's
33:43
over and keep on uh creating fantastic applications
33:49
with Dapper and happy coding my friends okay see you next time
33:56
[Music]