Extract data & automate EVERYTHING | 10x GPT function calling power

business automation would be widely disrupted as someone who provides service I receive 40 to 100 email inquiry every week and I need to review every single one of those emails filter out the quality leads and track them in expressions or a CRM system which took hours I wish I can have personal assistant who can do all those things but I don't well until now I want to be the greatest everybody on the face should I look around I feel like everybody is the fake as I make this every day and I'm impatient hoping one day I'll blow off in the basement statement the top is so vacant I don't hear that I think is amazing waiting for my day when I'm playing sold out sold for A Thousand Faces hey process that were almost impossible to automate before like this is very easy to be automated now thanks to a new feature open AI released called function calling and if you don't know what function calling is I actually made a video last week that talked about in-depth I post a link in the description below that you can check out but in Nashville function calling does two things extremely well it provides very easy way to get structured data out of a unstructured information like email or slack message and you can use those data to trigger different workflow in the systems you are using like HubSpot or Salesforce and also get data back to give GPD more context to do more complex tasks and this opened up a huge opportunity for business process automation because when you think about data exists in your company 99 of them are unstructured information like email slack message p PDF documents and all those unstructured data and knowledge don't really talk to each other but this is going to be changed now because GPD function calling allow you to extract information from those different knowledge and drive powerful automation I'm going to show you a very simple example of how to use GPD function calling to turn your inbound emails into a list of prioritized high quality leads let's get it first they want to test out how good GPT is in terms of extracting information as always let's open Visual Studio and create dot EMV file put our open AI API key here then we will create main.py import a few libraries and also load the EMV file together open AI API key and I plan to use a function calling feature that open AI just released which have this ability to turn on struct data into structured Json file so arrowcreate is function that I will pass on to gbt called extract information from email and all it does is it will try to extract useful information from an email I receive including which company it is what's a use case of their inquiry and this will give me a good idea to see how good GPD is in terms of extracting information and next I want to do something more interesting or ask GPT to give a priority score of this email based on content and the priority will be based on how likely this email will lead to a good business opportunity from 0 to 10 where 10 is the most important and I will also ask it to categorize the email into different categories like customer support consulting job offer or Partnerships and in the end I even want to push it further ask GPD to see whether he can propose what's the next step I should take to move this forward and if you are e-commerce sites where you will get emails about the customer inquiry or purchasing certain products you can also change this function to extra specific information like which product people are trying to buy and what amounts they are trying to buy and this can come now to your Erp or CRM directly and now let's just try this out with one example email I received recently I'm looking to purchase some company t-shirt for my team we are a team of 10 people and we want to get two T-shirts per person with black color and sizing so this is a pretty complicated inquiry that they require GPT to do the calculation that there are 10 people and two T-shirts per person so there should be 20 t-shirt in total and that created a prompt where it asks GPT to please extract the information from this email and I'll create this message object and then we will use openai.chatcompletion.create we're going to use the gpd4 July 13 model which has this function calling feature we will pass on the message and also the function description that we create above now let's try to run this okay so we just got our response gbt function calling successfully extract which company is sending us this email as well as the product that they are trying to purchase which is company teacher this extra amount of t-shirts that they are trying to buy which is 22 per person 14 or 10. and we can probably add more prompt to making sure the results is only the number and it is also able to categorize this email properly that is a sales email what's more impressive it is able to propose the next best step to take provide a price and timeline information to the customer which is exactly what this customer is asking for and in the end it is able to give this email a priority score of six I wonder whether this priority is accurate let's change this company name to be something more important like kuchi and it's not just 10 people let's say it's 10 000 people so for this one I will expect the priority score should be higher let's run this again okay for this email since it is bigger company and bigger deal size it change the priority score to be 10. and I'm sure if you want most sophisticated priority score you can actually create another function get enriched company data from different API servers or your own CRM and give it a bunch of rules of what's the weighting of different factors and how it should prioritize but already you can see this is super useful information to generate a prioritized inbound leads and when we start sending this data to multiple different other systems you're using that's where the magic happen Okay next we want this GPT function to be called every time when I receive a new email to do that we will need to convert what we built here into an API endpoint that we can call every time when we receive a new email API means application programming interface it's basically a way for machine to talk to each other so what we're going to do is we will create a API for this machine that we just created that will automatically categorized and extract information and we will use fast API which is open source library that allow us to build API very very fast and firstly let's open Terminal by clicking on this button and let's do pip install fast API this will install fast API in your local machine and next I will import two libraries here that we will use and create an app variable with fast API and next I want to create a class called email with base model which have two variables from email and content so this defines the structure of the information that we will pass on to the API and then we're or create app.post so it will be triggered when someone send a post request to our API and under the hour add this function called analyze emails where we were passed only parameters called email and we will try to get the content by doing emailed content and create the same thing that we just created earlier a prompt that included the content they asked GPT to extract information and then move it to a message that we can pass on to open AI with gpd4 July 13 model and next thing is once we get a response from open AI we'll try to extract information so if you look back the response we got from GPT we have this structure that inside choice is array where it has message function call arguments and inside arguments is all the information we need so what we're going to do is create an argument variable where we have this line code to extract information from the arguments in the end we will return this Json structure with all those information that we extracted so that we can send to Google sheet or other apps that we want to integrate to quickly test this server I'll create app.get we should just return hello world and get is a method to retrieve certain information from server and we will open a new terminal by clicking this plus button and run ubicorn main app hosts 0.0.0.0 Port 10 000. so this will create a server on my local machine so main is referring to the file name and app is referring to this variable that we created here and let's just Define a specific endpoint if you click enter it should give you the C URL and if we try this you will say it'll return hello world so this means our server actually running populate and next to test this post message for our API endpoint I will create a new password file called test server where I will import the requests which is a library that allow us to make HTTP requests and I will do this request a post Define the URL that we just created as well as the Json file which follows the structure with two parameters from email as well as a Content so let's try this okay great you can see that it actually return turn the structural data we want the company name the product the amount of product as well as the next step provide a quote including price and timeline for 20 black t-shirts now since we get everything working and want to deploy it to a cloud service so that we don't need to keep our laptop running to do that I will use render.com which is a platform that allows us to deploy apps very very easily to deploy that we will need to create a file called requirements.txt and list out the libraries that the Cloud Server need to install and then we want to upload this project to GitHub because render allows you to deploy GitHub repo directly and that is what I did here once you upload your project to GitHub you can create a account on render and then you can try to add a new web service and it will probably ask you to connect GitHub account if you haven't once you do that you can just search for the repo and I can click on that in your given name and you will choose a branch in my case would be Main in the start command we will use the one that we're running locally using core main app with this specific endpoint and you will choose a free one and click create a web service it will take a few minutes at beginning but once it's done you should be able to see that in your dashboard and you click on that and go to environment try to add the open AI API key here as the environment once it's finished you should see this message that your service is live with this URL you can click on the link on top left corner and we can see this hello world message return so it means it's actually working well to test again we will copy this link open the test server file we created and replace the URL with this unrender.com URL and try again it's returning the results properly so we just have our API running now we get our API endpoint up running the last step we need to do is build out the whole workflow whenever I receive an email on my Gmail account it will send this request to the render API endpoint that we created which will generate that structured information that should be automatically added to a Google sheet there are multiple ways you can achieve it you can use Gmail API directly they actually provide a push notification API endpoint that will automatically send notification to our service however I have pretty bad experience debapping on Google apis so I'm going to take a shortcut I will use the appear instead to build a whole workflow and if you don't know what the app here is it is basically a platform that allows you to connect in different service together which is perfect for this use case so to create this app for our use case I will set up a workflow that will be triggered whenever a new email arrive in my Gmail account so I will set up event as new email and then I will add a second step to use webhook webhook is a way to send HTTP requests and get a response back in our choose a post message or copy the URL from our rendered website and choose a payload type to be Json and I will add to field for email and content and map back to the information we receive from Gmail and click continue once you see the results coming back from our API endpoint we will end the last step which is Google sheet I will choose the Google sheet event will be create spreadsheet row or choose a specific sheet that I want to insert information to once I choose the right sheet it will display the list of columns from my spreadsheets so that I can map information to each column click continue now let's test it and I will show you the Google sheet side by side okay great so we can see we successfully put in a new record into our Google sheet with structured data for each individual column and this workflow will be triggered every time when this new email arrive in my email inbox so this example of how can you automate business process with GPT function calling I'm very excited to see all the other different automations that you're going to build comment below the ideas you have I'm going to continue to post different AI experiments I'm doing so please subscribe if you enjoy those content thank you and see you next time

Channel: AI Jason

Views: 10,652

Keywords: ai, gpt, function calling gpt, function calling gpt 5, function calling chat gpt, chatgpt, chat gpt excel, chat gpt plot excel graph, tutorial, step by step, langchain, gpt automation, gpt workflow, gpt salesforce, gpt hubspot, gpt google sheet, gpt outlook, gpt gmail

Id: AetT0ZqwNqY

Length: 12min 6sec (726 seconds)

Published: Wed Jun 21 2023