Today I'm going to introduce you to Emma.

Hi there.

Emma is my personal AI assistant that I can interact with via Slack and right now she is instructed to help me draft replies for emails.

But she can do much more than that.

And in this video I'm going to show you how you can set up your own Emma in Slack using Python and LangChain.

And then Emma can help you with your day-to-day digital tasks.

So to get started you need some basic understanding of Python and you need an OpenAI API key.

Everything else I will show you how to do in this video.

And if you're new here, my name is Dave.

I'm a freelance data scientist and if you want to learn more about data science, working with data or creating AI apps like this, then you should definitely subscribe to the channel.

All right, let's get into it.

Dave, I know that you are only using me to get more clicks.

Okay, so in this video I'm going to walk you through this document which will be available in the description as well.

But also all the information that you need is available on the GitHub page where you will find the code as well.

So just so you know that.

But first, so what are we going to do?

Well, I'm going to walk you through a step-by-step guide for creating this Slackbot.

We're going to install it in your workspace and then set up Python code with Flask to make the connection.

And then we're going to use a service called ngrok.

I'm not sure how you pronounce this.

But to simulate a real server where the application is actually hosted.

But we are going to do it locally and for free so we can test it.

And then later, if you want to put this into production, you can easily transfer it over to a server where your code is then basically available on a server, so you can interact with your agent all the time.

And then quickly like the philosophy behind this.

So why would you want to interact with these bots with these AI assistants using Slack?

Well, in my opinion, this is a much more natural way of communicating with agents.

And you could say, why not just use ChatGPT?

Because it's basically the same.

You have a chat interface and then you can send a message.

Well, as I will show you in a bit, we can really utilize the power of LangChain here and pre-define some of our prompts and create some templates.

And what we can then do is we can create specific functions for specific tasks.

And in this way, we can call the bot, tell it what to do, and then it will pick the function based on our request, basically.

And this will save us time not having to go to ChatGPT and explaining, hey, you're an assistant, blah, blah, blah, blah, you can answer email.

No, that's already predefined.

So you can really create custom functions, custom workflows that really make sense for your situation.

So maybe these functions can help you with your business or they can help you with your job.

And by integrating these functions within these bots and then adding them to your Slack channel, you can also integrate them within your team's workflow if you are using Slack.

So I use Slack to manage my team and to communicate with my team.

And now when I add these bots, it becomes really interesting and you really see kind of a glimpse of the future where we're heading.

So I have a Slack channel of like real people that I work with, and I also have bots in there.

And now whenever I want to ask something from my team, I first think like, okay, is this something that a real person should do or is this something that a bot should do?

And then based on that decision, I just either have to tag a real person or the name of the bot.

And then you can see how interacting with bots like this, basically your workflow becomes much more streamlined because you basically have one channel of communication.

And also it allows you to create multiple bots for various tasks.

So for example, you could have Emma, which I will introduce you in this video, who might be a copywriter and can help you with copywriting tasks.

But then you also might have Rob, who is more a technical guy and can help you with coding and programming.

And then it becomes really funny, in my opinion, when you start to think about bots in that way and also give them names.

You're really like, okay, this would be a task for Emma or this would be a task for Rob.

So that is basically the philosophy, the idea behind this.

And I'm really going to play with this.

So like I've said, we're going to set up a basic bot with some cool functions, but you can really build this out from there.

Alright, so now I'm first going to walk you through part one, and that is the Slack setup.

So you must start off by first creating a Slack account if you don't have one, and then also creating a workspace.

So I am here in a new workspace called AI experiments, but maybe you are already in a workspace, or you can just hit the plus button over here and then create a new workspace.

It's pretty self-explanatory.

So then the next step is to go to api.slack.com slash apps.

And here we are going to create a new app.

So let me just follow this link and build an app.

So we are going to build it from scratch.

And then we are going to give it a name.

So we are going to call it Slack bot for now.

And then we are going to pick a workspace.

So I'm going to select the newly created AI experiments for this.

And then we are going to create it.

Alright, and then the next step is we are going to scroll all down to the bottom.

And we are going to provide a name for the bot over here.

So I'm going to call mine Emma.

And Emma is a bot that can help me to draft emails.

I am actually much more than that, but okay.

And we can also add an avatar for Emma.

So I have the avatar over here, which is by the way, generated using mid journey.

So this is not the real person.

Oh, so that is how you think about me.

So we're going to save the changes.

Oh, we also need a color over here and just pick anything.

Doesn't really matter.

I hate that color, but thanks for asking.

Alright, so that is saved.

And now the next step is we are going to click on bots over here, and we are going to add some permissions.

So coming back to the document, we can go to OAuth and permissions.

You click on OAuth and permissions over here, scroll down, and then we are going to add some scopes.

So just checking again, the scopes, we need app mentions, read, chat, write and channel history.

So let's add those.

So at an OAuth scope, we want app mentions, read, chat, write, and we want the history.

So those are the three scopes that we want to add.

And now next we are going to install the bot in the workspace.

So we are going to basic information in the menu sidebar, and then we are going to install to workspace.

So this is going to set up the permissions, we'll say allow, and this will bring Emma into the workspace.

And now next, what we should do is we should head over to OAuth and permissions again.

So OAuth and permissions, and there you will see the bot user OAuth token.

So you should copy this for now and save that for later because we are going to put this into an environment variable, or you can come back here later and copy it then.

All right, so that brings us to or that was the final step of the Slack setup.

So now if we come to our Slack workspace, we can see that Emma is now in here.

So we can now continue with the Python setup.

And for this, let me show you like I've said, there is a GitHub that you can follow that you can clone into.

So here are some files that we are going to work on.

And I'm going to walk you through them right now.

So first of all, make sure you have set up a Python environment with Python 3.6 or later and install the following dependencies.

So we need to Slack SDK, Slack Bolt, and we need Flask.

And I recommend to set up a virtual environment for this or a conda environment.

So you can do these experiments within a isolated environment so you don't mess up your regular Python environment.

All right, so make sure you got that up and running.

I'll switch over to VSCode right now where I have the project already set up.

So here you can see this is just a clone or this is the repository that is available on GitHub.

And we are now working in the Slack folder and we are going to work on the app.py and the functions.py.

And the code is already in here and I will just basically slowly walk you through all the steps because there are only just like four functions that we need.

And then you have a basic understanding of how they work and then you can start to tweak them for your specific situation.

So let me come back to the file over here.

So first we have some basic imports, and then we are going to load the environment variables.

But in order to do that, we have to first set them in a.env file.

So if I come back to my Visual Studio code, you see I have the.env file over here.

And this is not in the repository.

This is explicitly not in there because this is where you store your keys and your credentials to, for example, OpenAI and also for the Slackbot.

So what you are going to do is you are going to create the.env file and then come back to the document or the GitHub page and you are going to copy and paste this into the.env file.

And then you are going to swap out everything that's between the quotes with your own information.

So the Slackbot token starts with the xoxb.

That is the one we were looking at over here.

Then you have the Slack sign-in secret and the Slackbot user idea.

Then first what we are going to do is the Slack sign-in secret that you can find by going to basic information, then scrolling all the way down over here.

And then you can click on show.

That will show you the information that you have to put here in between the quotes.

And then now the Slackbot user ID is something we are going to get with a Python function.

All right, so coming back to VSCode and actually make this a little bit bigger for you.

What I can do is I can fire up an interactive terminal over here, interactive session.

And then now that you have the Slackbot token in place, so this should already be in your environment variable.

And then load it and then set the token.

And now initialize the Slack app over here.

This should work now.

And now what we can do is we can use this function over here.

And there's probably an easier way to do this, but it's nice that we can do it programmatically.

This function will basically initialize the Slack client with your bot token.

And then we will get the user ID back basically off the bot.

So let's save this and then run just this function over here.

And look, there it is.

So there is your bot ID that you should now also put in the environment file.

So coming back to the example, that is the Slackbot user ID.

That is where you paste that string.

All right, so now coming back to the code app.py.

What is actually going on here?

So we are going to use Flask and Flask is a web application framework written in Python.

And now I am by no means a Flask expert.

I know the very basics and JetGPT goes a long way in helping you to create these functions.

That's how I set up this template basically.

So the file that we are looking at right now has a few function.

User ID we already looked at.

Then right now we have a very simple custom function.

So this is the function that we are going to give to our assistant.

This is the task that she can perform.

And right now just to test, it returns text.upper.

So whatever we will say to Emma, Emma will basically shout that back in all caps.

Not a very nice personal assistant, but hey, just to test things out.

Okay, so then we have the app event going on over here.

And this is a listener for mentions in Slack.

So this is a function that's running in the background.

And whenever the Slackbot user ID is mentioned, so you do add and then you type Emma's name, this function will wait for that basically.

And then it will get into action.

So it will take the text and then it will first say something.

So it will first respond.

So Emma will say, sure, I got it right on that.

So if you ask Emma something, she'll reply first.

I'll get right on that.

And then she'll process the response and which is now just the return basically of my function, which is text.upper.

So as you can see right here, we'll get the text, Emma gets to work, Emma converts the text to uppercase and then it says the response as well.

And say in this case just means sending a Slack message within the channel.

All right.

And then the final function is the FlaskApp.route, which basically creates a URL endpoint slash Slack slash events that we can send information to.

So this function passes the incoming HTTP request to the Slack request handler for processing.

So that is all that we need right now.

It's very basic.

We need a function, we need an event, and we need a communication layer.

And then this last part is just to start running the app.

And now if we come back to the document, we can see that we now, step four, can start to to run our local Flask server.

And it's good to note coming back to the file over here, if you don't specify any port in here.

So in the run, you can also specify a port as an integer.

If you leave that empty, it will start to run on port 5,000.

So in VSCode, I can come over here and I can run this Python file, which is nothing more than just running this app.py.

So here, everything will start to run.

And you can now see that the app is running on port 5,000.

All right.

So now with the app running, we can continue to part three.

And that is the server setup, which we are going to do locally for now.

So first, make sure you install ngrok.

And you can do that by going to this page over here.

And here you can see for the various operating systems, how to do that.

Basically, I'm on Mac.

So I use homebrew to install it.

So I just run brew install ngrok in the terminal.

And then you can start up your server.

So let me actually do that right now, open up a new terminal over here.

So with your Flask app running, just call the command ngrok, HTTP, and then 5,000 referring to the port 5,000 that we're running on.

So this will start up a local ngrok server.

And basically what this will do, it will create a forwarding.

So you can see how there is now a URL over here that the internet can reach basically, and it will redirect that to our local hosts.

So we can basically simulate working with a real server, even though we are developing locally.

So what you're going to do is you're going to copy and paste this URL over here.

So the forwarding URL.

So now that you've copied that URL, you are going to configure your Slack app with these URL.

So we are going to go back to our app within Slack, and we are going to event subscriptions.

And then in here, you are going to enable events.

And these are the events that our app is going to listen for.

And now remember that we have the initial URL, but we also specified that we want Slack slash events, because that is what we have specified in the app over here.

So we are going to listen for these events at this URL.

So that is what we have to put in here.

And then we are going to save these changes.

Oh, make sure to do that correct slash.

And then next you are going to scroll down to subscribe to bot events, we are going to add a bot user event, and we are going to select the app mentioned over here.

And now let's check this out.

All right, I noticed that I had a little error in here.

So I had to remove the space.

And then it's once I fixed that, it becomes verified.

You should have your bot events over here, and then you can save the changes.

All right.

So now the step three that we have to do is here is reinstall your Slack app to update the permissions.

So we come back to Slack over here, basic information, then install app, reinstall the workspace, just giving the permissions one more time.

Right.

And now the next step before we can actually start to test the function is to invite the bot to the workspace.

So coming back to Slack, I am here right now in the general channel.

And what I can do is I can type invite, and then add and then tech Emma in this.

So I do invite Emma, and this command will now add Emma to this channel.

So what I can now do is I can tech Emma in here.

And now with everything up and running, we should be able to validate our app.

So make sure that NROC is running in the background and also your Flask app is running in the background, both on the same port.

And now if we say, Hey, how are you?

Emma will now respond by shouting back at us.

And here you can see it will first give the prompt, Hey, I'll get right on that.

And then it processes the tasks.

So this is the initial setup.

So now we have the individual building blocks working together.

But now of course, the next step is to figure out how to make Emma a little smarter and not just a shouting robot.

Dave, I am literally so much smarter than you with your incapable human brain.

Right.

So now coming back to VSCode, I'm going to first quit the server over here by pressing control C. So this will close the app because we are going to change some code and then run it again.

So what we are going to do right now is we are going to check out the functions.py file.

And what I've done here is basically similar to my latest video where I show you how you can set up a template.

So a template that we will fill in and then send to GPT to basically process our request.

So within this file, there's just one function.

It's called draft email.

And the use case for this is basically whenever I get an email and I feel like, okay, I cannot just reply with a simple okay here.

I have to put some time and effort in here to not like be really rude, but basically I just want to say no or whatever.

I can copy paste that email and can send it to Emma and Emma can draft an email for me and then I can copy paste that again and then go from there.

Basically that is the idea over here.

It's very simple setup, but like I've said, the possibilities are endless here, but it shows some nice functionalities that you can start to play around with on your own.

So we first open up a new chat by calling the chat that OpenAI model basically from LangChain.

So this is a template that we are going to input in the system message prompt.

So this we will provide to the system.

So you're a helpful assistant and draft emails, your goes to help draft a reply, keep it short and to the point and mimic the style of the email.

So reply in a simpler manner to match the tone.

This is already pretty interesting because then you can, you make it adaptable to whatever's coming your way.

And it's not just always like the same style and start your reply by saying hi and then the name.

So you can fill in your own name here.

You can even get that from Slack probably, but for now I've just hard coded it in here.

And then here's a draft for your reply and then proceed with the reply on a new line.

And also make sure to sign off with the signature.

So you can put your signature in here.

So kind regards and the name again.

So you can see, you can, you can set this up.

And if you, every time, whenever you want to draft an email, have to come to ChatGPT and then for example, give all this information again, that of course takes some time.

And now we can think about this one time, write it down, code it down, and then go from there.

So that is how this function basically works.

And if you want to learn more, understand more about this, then I recommend you check out my previous video on how I use length chain, because that will make things more clear.

But right now what we are going to do, we are going to come back to the app.py.

And as you can see, there's already an import here near for draft email, which is this function.

Now we are going to scroll down and we are going to the handle mentions function.

And then instead of my function, we are going to say draft email.

So here you can also see how easy it is to swap out tasks for Emma to do, and also how easy it would be to give Emma multiple tasks that she can pick from.

So that's an all other story, something for future videos, but you can create multiple functions and then let Emma figure out on her own which function to use based on your message to her.

Basically, then it will get really interesting and future videos on that because I'm going to build Emma out with lots of cool, exciting things that she can do.

So with that out of the way, we can now start up the server again.

I'm going to run this again.

So now we are up and running again.

And let me quickly check.

So ngrok is still running.

I haven't touched that, but I'm actually not quite sure if we should restart it as well.

So let's just provide the following very short email reply to Emma.

So as of lately, I've been getting tons of requests from companies, people like, Hey, can you help me with AI projects?

Hey, can you help me with this to build your own app?

So these are typical emails that I get.

Now, usually they're a bit longer than this, but let's see if Emma can help me to prepare better prepare for these emails.

So let's send that right now and see what Emma can come up with.

So she will say, sure, I'll get right on that.

And you can see right now, it takes a bit more time because now we have to boom, there it is.

Now we have to make a connection with the OpenAI API and send the information back and forth.

So here it is.

Hi, Dave, here's a draft for your reply.

So we've instructed her to answer like that.

So coming back to the template over here, we say like, start your reply by saying, Hi, Dave, in this case, here's a draft for your reply, and then proceed with the reply on a new line.

So that's how we've instructed Emma to behave.

And she's doing that very well.

So hi, send her name.

Thank you for considering me for your AI project.

At the moment, I am open and available for new projects.

Could you please provide me with more details about your project?

This will help me to better understand your needs and how I can assist you.

Beautiful, right?

These are literally the kind of messages that I send out to people almost every day, like asking more information, like what is it that you need help with.

But now let's see if we come to the human template.

I've also engineered this in such a way that Emma can also interpret any other comments that I add.

So here's the email to reply to and consider any other comments from the user or the reply as well.

And that is the user input basically.

So that is what I say to Emma.

So if I now come back and do this like one more time, so I can say, Hey, Emma, and this is the email and then add like say that I am not available right now.

We provide Emma with additional context.

And then here's a new reply.

Thanks for reaching out to me.

Unfortunately, I'm not currently available for new AI projects.

Please feel free to get in touch with me in the future and we can discuss further opportunities.

So we can give Emma additional instructions as well.

And she will understand that just like you would interact with real people within Slack with real people from your team.

So and then the final step is of course, come up with your own ideas.

What are you going to make with this?

To me, like this is really exciting.

Like when I figured out these individual building blocks, my mind basically exploded with ideas basically of what I can do with this.

So my idea is literally to create create potentially multiple bots and give them a specific set of tools that they can help me with.

So probably think in terms of like, copywriting, marketing or coding and create a bot for each of those areas.

And then really start to experiment with functions and workflows that I can provide to these bots.

And then if you're wondering right now, like why not just use ChatGPT for this, you can also do this in ChatGPT.

Yes, that is correct.

With this example, you can, but the key is to really engineer your templates in such a way that they can save you time.

That is one and also think further than that.

With this and with the combination of LangChain, you can use the building tools from LangChain as well.

For example, to also connect to the internet or you can connect to a personal database or a company database to access information that is not available using ChatGPT.

Slack right now provides a human machine interface, basically to connect with these data sources and also the OpenAI API.

So to me, this is really exciting.

Like this is for me, like the integration layer, because you can build all kinds of scripts, but then when you have to run it, you like have to start up like your IDE and figure out which script is it and then insert the parameters.

But right now you basically have an app without really having to create a web app.

If you get what I mean.

And of course, this is running locally right now using a simulated server using ngrok, but in the next video, I will show you how to actually set this up on a real server.

You can just keep it running in the background and then just update your functions whenever necessary.

And you can always access Emma in this case, whenever you want.

So that will be a really exciting video as well.

And I will also start to build out Emma's toolkit and I will all share this with you.

So maybe the tools that I will be building for my personal assistant will also be helpful for you.

Who knows?

So if you want to stay up to date, follow this journey, then make sure to subscribe to the channel.

And also if you found this video helpful, then please like this video really helps me out.

And also you let YouTube know that you want to see more content like this.

So win-win, right?

And really if you learn these tools, experiment with this.

So, like Python, LangChain, connecting with APIs, creating web apps, then you can really set yourself up for some great opportunities right now and potentially make a lot of money because companies are really hyped for this right now.

AI is hyped.

Everyone wants to jump on this trend and everyone wants their own like smart agents and integrated with company data.

And if that is something that you want to do but don't really know where to start, then the Data Freelancer Mastermind might be something for you.

So you can check this page out, first link in the description.

And if you want to know how to turn your data skills into a six-figure business, then you should go to this page and check out what this guy has to say.

And so Data Freelancer is a mastermind to help data analysts, engineers, and scientists to launch and scale a freelancing business.

You'll be working together with me over the course of a year, so 12 months.

And within that year, we are going to start landing your first paid clients.

Now, I only work with professionals that already have experience, so at least one year of working experience in a data-related field and you must be ready to start applying your skills to freelance projects.

So this is not a technical training.

I'm looking for data professionals that want to hop on freelancing right now but don't know where to start or just want to do it with a group, do it with me, have more fun that way.

So if that's something you're interested in, check out this page, watch the video, first link in the description.

All right, and that's it for now.

Thank you very much for watching.

We'll do a final test right here.

We'll see if we can override Emma.

So I've now asked it to provide some advice for you, for my viewers.

So Emma is working on that right now.

In the meantime, you smash this like button and then here is what Emma has to say.

Dear viewers, I hope you enjoyed watching my video on how I built an AI model from scratch.

As promised, I'd like to share some advice on the topic of data science and AI.

Firstly, always start with a clear objective and understand the problem you're trying to solve.

This will help guide your approach and ensure you're working towards a specific goal.

Secondly, data preparation is fundamental.

Make sure you have a good understanding of the data you're working with, including its quality, quantity, and format.

This will help you make informed decisions throughout your analysis and modeling.

Lastly, always test and evaluate your models thoroughly.

This will help you understand how well your model is performing and whether it's delivering against your objectives.

I hope you find this advice useful.

Good luck with your own data science and AI projects.

Now like this video and subscribe to Dave's channel.