z9aP0_Qqq2A
Let's build with Zapier and AI (100+ videos)
Transcript
[00:00] how do we use assistance API with absolutely no code and start answering emails for customer service let's find out the idea of this tutorial is I'm going to show you step by step how to build out an assistance API that we can use and access with no code set up automations for and then at the end give you a pragmatic way to start applying this in your business welcome back to Corbin AI y'all where I'm showing you how to start leveraging artificial intelligence whether that's just for your business or just going over cool stuff in the news check me out on Twitter here real quick if you want to get updated on new vide that are coming
[00:30] out you know kind of alluding to it also if you've been following me I've been saying this Swedish House Mafia plus coffe is Max productivity I this is I should probably be embarrassed that this says 0.5% fan but is what it is let's jump into today's tutorial and learn how we can start Leever leveraging assistance API so if you don't know what assistance API is and this is the first time you're seeing a video like this go Ahad and check out that video right there as I go over like just the capabilities of assistance API but just so you know what're we're diving in today so we're going to use assistance
[01:01] API here and we're going to use zap so I'm going to link two videos off this side here one that just shows you how to integrate zapier with open Ai and the other one just showing you what event assistance API is so that in this video if you do decide to come back you you know have more context let me move my mic a little bit closer here okay so we're going to create this actually in open AI front end here and then we're going to proceed to use it and zap yours backend so the purpose of today's video is we're going to create a customer service type bot or a assistant that is trained on our businesses data and we're
[01:32] going to go over through the processes of how do we even get the type of business data the best way to format our business data and stuff of this nature so to start off here we're just going to call this customer service and we'll add a little Dash here called Web Cafe so your instructions be very uh specific here so for us we're just going to say you are a customer support agent for web Cafe AI trained on our business data so the reason I'm
[02:03] doing this one sentence here is that as you'll see in our zap your automations we're g to actually be able to provide additional instructions so that's what makes theis API so cool one we can actually save chats but two we can actually build on top of cont we can build on top of instructions for specific use cases and for specific context this will make more sense as we dive into here also I want to point out by the way as well this is extremely powerful so I'm making this for customer service but man you could do this for a ton of stuff when it comes to assistance
[02:34] API I can make one trained to make ebooks I can make one trained for XYZ reason and loaded of data but for now we're just going to go ahead and do this we're going to use a model of gbt 4 1106 preview as you may or may not know they've under they slashed their prices so it's 66% cheaper actually 67% cheaper than the gbt 4 model that were we've been used to up to this point for here though for our specific context we're going to do retrieval and retrieval is going to allow us to add files big deal here so we can add a good amount of
[03:04] files here but what is very important for you to do with these files is we need to format them correctly we need to use a TX txt why are we doing that we're doing that because one it compresses the data so it's smaller file size but two alternatively rather than me loading up a really big PDF that maybe has superflous information let's just load up what we care about and if it looks you know aesthetically does not look good it doesn't matter because we're
[03:36] just feeding it to a bot that bot's going to read it and that's all that matters because the end conversion point is just so it's train knowing this let's gohe and walk through step bystep how we're going to do it for our personal Marketplace here called Web Cafe software we sell a lot of cool stuff here if you want to check it out but what we're going to do here is we're going to go ahead and grab relevant information that I care about and I want to train my customer service bot on to start off here let's just go to the FAQ page now whether you're an agency that's working on behalf of a business or you're a business yourself you will know where the readily available information
[04:06] that you care about when it comes to the context to customer service will exist EG pass email chains EG pass conversations you had in DMS what were common questions what were common occurrences I mean you could have trained this to be a sales bot as well so maybe you load of sales data for like you know trying to convert individuals on email so for our context though I'm going to show you the most minimal or way to approach it that's kind of just like let's just load it into a new file and and just have it all in one file so for us we're going to do it by grabbing Links of our website that has relevant
[04:37] information that an individual might ask about so obviously the big one is going to be the FAQ to start so we're going to go ahead I'm going to link this in the description down below it's basically what it does is that it will take website links and convert it into text that is because we're going to be creating a text file together so I'm going to hit command V here I am not a robot yet um One Time Imagine in the future these videos are just coming out and it's just like wait is that is that actually a Corbin AI or is that actually Corbin we don't know indistinguishable
[05:07] so what it does obviously is it outputs it like not super attractive but what we're going to do here is we're going to grab the relevant information that we care about so for us we just care about the FAQ section so we're going to copy this text let's first make a new text file so I'm not sure what it's called on Windows but for Mac it's called text edit make a new document here we're going to jump back over to this website and we're just going to grab our relevant information so for us we're going to start off with the FAQ section here so I'm going to hit command C command V all right perfect so we have our FAQ section so far and then I'm
[05:38] going to go ahead and make sure I move myself over here so we're not making too much of a Ruckus shrink myself down and this add some more pages here so I'm going to go back to the FAQ what else do we care about here okay let's go to the about and I'm going to go and copy that go ahead and copy this link here I'm going to come back over to this site command a command V so in your context you would be providing relevant information about your business that you care about is that going to be I mean this also can go farther than that imagine we could provide this in the context of
[06:09] training uh employees that come into it so we provide some type of conversational assistant API that's like hey what if a customer asks me this and then it gives a response like there there this could be an internal use rather than external use as well knowing that let's go ahead and just keep adding stuff here so this is from our about page so about us provide relevant information let's see if there's anything else IID want to add here okay so the last page I'm going add here is going to be our licenses page this is in the context if you buy software from us we have two different types of license and this is a relevant question we may
[06:39] get asked a lot therefore I'm going to go ahead and make sure we get all this relevant information okay so now that we have all the information here and look at this I can just okay now that we have all this relevant information here yeah it doesn't look amazing aesthetically it doesn't look amazing but who cares end of the day AI is going to read it and all they really care about is the text so that is why we're going to do it in in this type of format therefore if you have a ton of information you want to load onto it this is how we structure it so I'm going to go ahead and save this so right now I don't know why and I'm going to go ahead and jump back over here uh Mac for some reason doesn't
[07:12] allow us to save as txt they going therefore I'm going to use a website that's called Cloud convert I'm going to upload the r RTF here and then convert it to a text file and the reason I am converting an RTF to a text.txt is because of the fact that that's even less data associated with it so you probably got that's probably I probably hammered that home too much but the name of the game is the less data you have but the type of data you provide is important when providing data for models okay so then once we have this data here I'm going to go ahead and add the file
[07:43] then let's go ahead and hit save perfect so we went ahead and have successfully created our assistant API here we got the instructions we got the assistant ID which will be very relevant for us in the future here's something really cool y'all we so we don't have to jump right into uh zapier to test whether this is working how we want it to work we can actually use a feature within open AI called The Playground feature no we're not jumping on monkey bars here this feature is going to allow us to see basically gut check the API system to see if it's actually trained on the data
[08:13] if the responses are up to part how you want the responses to be EG the instructions do I want it to be funny do I want it to have humor do I want it to be boring do I want it to structure in in bullet points whatever may be that's to where you put it here from here though we're just going to say this is our gut check what kind of licenses do y'all offer so I'm going hit add and run here coming back over here we have an individual business license and we have a commercial license therefore if this is true this should bring out an output from that text file we have right here that says that so let's go ahead
[08:44] and see what it comes up with I want to point out as well that it seems like internally when using the playground is a lot slower than what I've seen when using it on zaper or in a code context there we go so as you see we got the individual business license and the commercial license all the relevant information about those license that we provided in the uh data file on top of that more contextual information in regards to this really good stuff let's take this API and push it towards our zap your flows and create a pragmatic use case in the context of maybe an
[09:15] individual or potential lead emails you and you want to set up a draft email trained on the data therefore you can kind of look at it and then hit the final send in order to do so we're going to come over to zap your yall we're going to hit create new zap and use case here is going to be very simple so we're just going to do trigger we say uh Gmail and there we go so we're going to go ahead and do an event of new email I want to point out as well though if you say new email matching search so for example with our Marketplace we use Shopify are back in and a very common
[09:48] way that it comes in datawise is like a new customer message therefore if you want to mitigate honestly let's just do it that way so I can really show yall this is so it's going to mitigate and you don't overuse your automation typically when you receive leads you can e either use sediment analysis by chat gbt or text found within the email to identify whether it should go down the automation or not so in the context that maybe you run a Shopify store we are used to providing or receiving emails with the headline new customer message therefore let's just go ahead and show you how to do that real quick so I'm going to go ahead and choose the account
[10:18] here this is going to be our courses account and I'm going to send a fake email to it just so we can have some test data here uh search string will be like new customer message so in Theory I'd probably put in as much dictation as possible that's usually correlated with these type of emails but for now I'm going to go ahead and just send me a fake email so if you don't have test data yet this is really good I'm just going to send this to myself I'm going to talk to myself and as you see we got our test data here what we can do here and I'm going to go rename this to uh API in
[10:49] action even though everything's an API call in theory we're going test trigger here and we should see that as some test data here we don't see as test data so let me try again now we see us test data perfect just put a simple body here hey I was wondering for your software Marketplace what are the licenses I could choose from we're going to say continuous selected record here and this is where the magic happens y'all we are going to leverage the assistant API this is where it's really cool so instead of conversation we're going to do conversation with assistant and what's
[11:19] so great about this is we don't have to train it at all why because we trained it here so we've already trained it in our assistance you know situation here it's already been trained on our data it already has instructions on how to respond therefore we can call upon its assistant ID and the assistant ID itself doesn't have to be a variable rather it can be fixed text because the assistant itself can be used across different automations be you know it's been centralized to this one type of trained model of Chad
[11:51] gbt knowing that for this we don't have to search for the assistant we don't have to do any of this okay I guess in theory you could just click it here that that makes a lot easier but in theory I could hit custom here paste in here and be good to go but for the sake of this video Let's just hit customer service as we know anytime we're dealing with customer service flows in our automations now we have an API assistant that is dedicated towards these kind of conversations what this really helps everyone to do is it allows you to basically not need to prompt it as much you don't need to structure your prompts as good as you used to do it because if
[12:23] you can make one really good assistant API for specific context now I don't have to add as much text in this message but we can still add text and that's what's so cool there's a layering effect incurring here so we're going to say based on this email uh email title email body notice the first major thing if you've been if you've been watching this channel for a while now you're like wait we messed up we didn't add a context block we don't need contacts anymore
[12:53] y'all because it is the assistant API no contacts needed so from here here we're going to go ahead and add because it has a context from the the back end right uh set the subject new customer message at the body and you might be asking yourself why do you have parenthesis around it that is to ensure that the data is just like it doesn't overlap of other data points gbt just likes to read it better like that so we're going to say this
[13:24] generate also want to point out I could have made the assistant API really specific to the context of emails but for now we're just going to leave it where we have to kind of provide that structuring here we say generate a email response body to this inquiry format uh Max of four
[13:54] sentences at uh the end and sign off with so I'm going just show you a couple things here sign off with best regards actually we don't need that we can actually add that as fixed text don't do that Max is four sentences let's actually just do that for now what I like to do sometimes is limited prompting just to see kind of feel it out where Chad gbt is thinking from here though we can go ahead and hit continue here let's test this step and see what
[14:25] it comes out to be this is good this happened live I want to show you this so it response came it was confident but it was wrong why was it confident why was it wrong because I accidentally didn't add something that is pertinent here which is existing files it was the data that we uploaded to this individual already so we're going to have to do the web Cafe data text that's what I called mine now I'm going to go ahead and continue here and retest this step just to send that home that was this data text that's what we created earlier in the video here we have to reference it in zapier as well so it seems like
[14:55] basically that means that assistant apis can choose not to read the dat if you choose to do so so I played round with a little bit it actually looks like we need to add a little bit more prompt structuring than I expected one thing I didn't expect was basically we added the existing file we added I added the tools just in case to make sure that it has it enabled to retrieve set existing file but finally you got to add a very specific line in your message which says reference the underlying file for information on the business that seemed to be pertinent as that basically told
[15:25] Chad gbt hey don't give an answer look at the data and then give an answer then as you see from this response here we have the individual business license referenced earlier and the commercial license referenced earlier therefore here's what we can do that's really cool we can add another Gmail block here with this Gmail block we can go ahead and do something such as setting up a reply or create a draft so for this draft I can go continue continue hit the subject line let's just put in
[15:57] uh new customer message draft and then for the two we can go ahead and choose where the emote was coming from here so the from email if I can type in from from email okay and coming down here we can go ahead and put our body which would be the output found here so one thing that I didn't necessarily like was the end point there so what I'm going to do is I'm going to add a little bit more text here if necessary also we
[16:27] can add signatures labels attachments we have lead PDFs stuff of this nature what I'm going to do here is one last thing here we going to say generate an email response body to this inquiry just the body no text before or after no uh signature or name at the end see if that is sufficient here no text before and after if you're familiar with this channel you definitely know what that is I'm going hit retest here that basically just makes it so sometimes gbt wants to
[16:57] talk too much no just give me a specific data point here that's all I care about and here we go so I'm going to go ahead and keep prompting until I get rid of this part perfect so I got it to the point where just as best regards from here all I need to do to make my life easier is if you have a standard way to responding to this maybe it's your I could have honestly added it to the prompt as well how to end a email but for here best regards Corbin now let's do Corbin AI I'm going to continue here I'm going to send this to myself and
[17:27] let's see what it does there we go y'all so we got the draft new customer draft response to the individual's name gives us actual response based off our business data let me go and zoom in y'all gives us an actual response on our business data and provides it as a draft so I can kind of look over it okay that's good to go send it also keep in mind y'all we can train the instructions better we can add more context to the underlying messeng here this all layers on top of each other it's like a mult it's like a funnel cake it all layers on top of each other therefore this whole
[17:59] process can be optimized a lot further maybe you want bullet points in your email maybe you want attachments in your email maybe there is a certain sentence you want in your email EEG add the fix text whatever it may be we have just learned today how to start automating our customer service or at least save us like 90% of the time therefore if you feel like you learn something up to this point or at the end of this video make sure to leave a like it's completely free and help me here if you want to see more videos like this when we get to more you know automations for your business and productivity I'm going to leave a playlist at the end here we're diving into a ton of stuff when it comes to artificial Ence and zapier and how to
[18:29] start leveraging it you can check that out here also on this channel I do other stuff so I do like AI news there a really funny video that I did yesterday looking at AI videos which was like or a new technology called P art well they existed but they did an update uh really cool really cool so cool that like they got to chill out maybe a little too fast there we do other content as well where we're just diving into how to leverage language learning models EG the chat gbt or all those gbts you see down there below but without further Ado I'll see you in the next video thanks for tuning
[19:00] in and yes surprise I'm an AI Avatar make sure to explore more here at Corbin AI where we demystify AI for your personal and business life until next time