Collaborate on Duplicate Invoice Example for Higher Education with Ease Using airSlate SignNow

Duplicate invoice example for Higher Education

[Music] thank you Hello friends welcome to my latest blog about automating supply chain task which no human assistant dare to tackle in the very first video I have covered how to download and use llma2 models and model bits on Ubuntu machine second video I cover related installation documentation about how to do the same thing in a Windows environment I've also shown a use case about using large language models to monitor employee times and expenses in our organization in next video I covered rather an interesting use case I've been working on using large language models as an OCR and vision AI for documents in this video today we are going to put it all together and showcase another very specific but very useful use case we will see how to automate supply chain daily routine tasks which are very tedious boring and require extensive manual hours also want to mention that please see that this video is not about coding and rather it is just a high level design review session to discuss approaches to handle supply chain tasks now some of these examples use cases include how to handle three-way matching process like matching purchase order to invoices and received before approving payments and how to locate duplicate invoices when you are working with millions and millions of documents now I want to call out that all the links are included in the video description below like I said this video is not about coding let's discuss some high level design this is a typical supply chain data flow diagram most of the time a customer make purchases or give orders which include products which obviously belong to some category so product category is product catalog and product has all the product information like description and prices similarly this is another process flow diagram from sourcing process perspective and we're going to discuss all this in later today so for now let's assume you know you have a table so in this invoice table you have a purchase order and so for only given date a process app purchase order was generated and it has all the information like customer ID vendor ID and how many quantities of product or order so similarly we have another table called receipt so for the purchase order which have been generated you keep a track of receipt so for example if a product has been received in the warehouse or not so this another sample table we will be working with in later part of the video today similarly you will find tons of code like this in Notebook which generates supply chain data the whole point of doing this is that we want to learn and derive generic behavior from these millions and millions of rows and use this information in our language in our language models so for example given a production is shipped from a vendor which has a 97 probability of receiving on time and invoices were always clean there was no anomalies find so that means it's a good vendor now similarly so using this Behavior we want to locate bad examples in our data and we that information to large language models and filter them to stop payments so that's the whole point of doing this exercise and you know um the whole point of doing this three-way match process here similarly I want to State another example a lot of time what happens a vendor charges you suppose you are receiving billions and billions of invoices a lot of time Earth happens a vendor charges you it could be a mistake duplicate invoices so it's often a pain to figure it out that whether a receipt or invoice or expense or is shipped it's been a duplicate of another or not so we are going to discuss the approaches how to identify the duplicate advices received in your system all right so that's what we are going to cover in this very specific use case today so if you are not interested in supply chain I want to warn you please feel free to skip this video all right so now if you're still watching this I assume that you have interest in supply chain automation so let me introduce myself my name is Amit Shukla and I train neural network on finance and supply chain data if you haven't already please subscribe to this channel all right so we are going to talk about installation which we have already covered in previous video so let's talk about the typical supply chain business process if you work for some very large organizations you might be familiar with these terms these are the business process very high level business process and depending on your organization there could be dozens or more with such business processes so this is a typical diagram of your supply chain and you can have lot of different tables but they're very high level less you know let's assume that you have a customer address order product category and product and also order Indian voices Also let's say product category is your product catalog and products has all the product information like description prices and whether this was in contract or not and Order similarly has the detail of the customer and there and their Associated product orders so similarly on the back end there's a business process flow so typically what you do when from the organization perspective when you make this kind of purchases you buy the goods from your manufacturers and you buy those in like you know brilliant bulk quantity and you know so whenever you issue a purchase order you issue to a vendor and upon successful receipt of the product you also get charged through an invoice and a typical three-way match process is that you match what was ordered what was an invoice and what has been received and everything matches then you pay to the vendor all right let's do a detailed Code walkthrough here I have already covered a snippet of this python code in my previous video so here I'm using these libraries so for example r pullers and you know a couple of other python libraries to generate this data so as you can see uh first let's load some data that data what we have discussed like customer vendor product and product category again you will find details of this code in the video description below this is very very easy code I'm not going to work through this again I have in my last video I have covered this in extensive detail how to generate this kind of uh this kind of data so for example here I'm generating a customer data frame as you can see there is a customer ID name phone number address email ID and you can be very creative please use this script to generate as many lines of your records as you want to be so here in this so let's go do the shape as you can see I have 1000 records for Associated to this one to you know 1000 different customers so similarly let's create another data frame so here let's say vendor so vendor could be the same thing I'm using a vendor ID name phone number address and email ID so again similarly let's create 1000 different vendors here similarly I'm going to create product and here you know I'm using another script which creates 1000 different products here and again I'm just initiating them giving them a price now let's go create a purchase order so what is the purchase order purchase order actually is consists of all those things what we just discussed purchase order so for example on a given date this particular customer buys this this particular vendors product and in this much quantity so that's what you're seeing in this purchase order data frame here here the only difference is as you can see I've generated 100 000 different lines and you can be creative I personally and this particular you know instance I played with the 10 millions or maybe 100 million rows here all right so for this case I'm just going to work on the hundred thousand rows here so as you can see there are hundred thousand different purchase order have created now let's go create 100 000 invoices for those hundred thousand purchases order what we just created please feel free to use you know you can and play with the sample size here and you can manipulate the data and generate as much data at fits into your computer memory all right so here you are looking at the hundred thousand different invoices generated for those particular purchaser so as you can see I have created some data so where it doesn't match intentionally so I want to create some bad invoices uh just to you know so that I can locate those bad invoices and you know I can use those large language models to figure it out what are the battery devices what is what are the good invoices similarly safety status so you can see like you know not so for example I have hundred thousand different purchase order not all of them have been received so this is again data I have created and as you can see there is a extra fields on top of this purchase order now then once you have imagine you have all this data coming from your organizations then you want to Define you want to you know derive a generate Behavior buy this product what are the chances that what are the average quantity ordered what are the average price I have paid for this product over the period of the time so once you have built this kind of General statistics you can use this historical statistical information pass it to your large language models through prompts and then you can derive the conclusion that okay this on this product if I get it from this vendor what are the probability that this vendor always pays beyond the time so the whole point of showing this you know these data sets and this building prompts and these things to you that we want to you know use the make use of Lord language models and we are just passing the statistical information to those llms plus please remember more the more statistical information just like to human brain so for example if you make a decision more statistical accurate information you have more facts you have you are going to you know make better decision so that's exactly what I'm doing using the llms of here right so now then you take one bad invoice and you take that particular invoice you pass that data to the LL alignment saying hey my generic Behavior was overall for that product from that vendor it is always good so for this particular instance this for this product that wonder is charging with this engine mount is it okay to pay that or not so again I have you know I've included some sample code here where you can call the chat GPT or you can call the meta llma models here now again I'm not going to go through the details of this code uh please visit this link so I'm going to include this intro this video description below so likewise for the next section duplicated devices I have these are the steps we have followed again you are going to find I am going to include the link to this notebook here and you will find code how to you know what is the approaches to look at the duplicate invoices I'm not going to cover everything in this video today because these are just the samples and I can understand like you know it could be quite cumbersome to cover these kind of details but you will find all the source code and everything included in the video description so again this is a very specific use course I realize it's not applicable to everyone it's only for the supply chain people who understand like you know who want to automate the supply chain or supply chain business process all right so now please go through this you know code and if you have any issues please feel free to open an issue at my GitHub repository and I'll be happy to help you out thank you for watching