Invoice Reconciler - Built with LlamaCloud

Key Takeaways

In this video, let's take a look at one of our demo applications, the invoice reconciler. And all of this is really about having a AI agent that is able to look at your contracts, look at your invoices, [clears throat] and try to reconcile them, deciding whether the invoice matches the contract agreement you have, whether there's anything that should be corrected, whether there's anything wrong given the agreement you made. So this application, as you can see, we have a dashboard, but really the most important three sections here are the contracts, [music] invoices, and reconcile. Let's take a look at contracts. As you can see here, we have a bunch of contracts. And what's happening in the background is every time you upload a new contract or an invoice for that matter, it gets [music] uploaded to your index in Lancloud. So we can have a look here for this application. We have two indexes contracts and invoices and we have six contracts in place and we have seven invoices. Similarly, we also index a bunch of invoices. And one thing that you can start by doing is both for contracts and invoices, you can use this search bar as sort of like a chaff application with your invoices. For example, what have we bought? And this will act like a simple rag application looking at all of the invoices we have and it will site the sources and give me a summary of what I have bought. And similarly you can do the same thing with contracts. You can for example ask what kind of contra contracts do we have in place. Let's start by playing around with this. And let's see here if we have a look we have seven invoices. And maybe we can start by adding yet another invoice. And the first thing you'll see happening is that the application will say that it the invoice is being processed. What's happening here is that we're actually sending the PDF to Llama Pause. And Llama Pass is making sure that it's converting the contents of the invoice into an LLM readable format such as markdown. Once that's complete, we can have a look at the contents of the invoice now in Markdown. And I can say save and sync one file. And you can see that that's getting indexed in llama cloud in our invoices index currently. And also in the application you can see that the indexing is in progress for this particular invoice. While that's still getting indexed, the next thing we can look at is the reconciliation. So we have two important sections here. one for reconciliation and the other for you to actually define the rules for that reconciliation. For example, here we've defined a set of rules about what an invoice should contain or what should be looked out for in both contracts and invoices to make sure that the invoices don't have anything missing etc. So here we've defined rules such as correct math, unit price match, line item match and so on. You can also additionally add your own rules here. You can also describe what those rules mean so that the AI agent understands what it should be looking for. All of this is combined in a llama agent workflow so that the workflow is both aware of the contracts, the invoices, the rules in place, and it also comes with a prompt explaining what the agent should be looking out for and how it should be doing the reconciliation. In the reconcile tab, we have a few more things that we can do for invoices that we've uploaded. You can either view the results of the reconciliation or for anything that has not been reconciled yet, you can go ahead and hit reconcile. So for example, we have one here that is still pending from before and we can hit reconcile and the agent will start analyzing based on these rules. So you can see that this particular invoice was rejected. If we expand this, we can see the reason why it's rejected. It says the invoice violates three business rules. The invoice clearly identifies a recognized supplier and has an invoice date that fits within the contract period but is missing vendor address information. Prevents an exact this prevents an exact match and so on and so on. If you want to see more details as to which rules were rejected and why, you can scroll down and have a look at them as well as seeing which rules have passed and with what confidence. Once you're happy with this, if you agree, you can save the result and it will get saved to the overall reconciliations dashboard. Other than this, you can also see all of the previous reconciliations. You can have a look at the invoices and also view the results of the reconciliation. So here we can see that this particular invoice was approved and [music] as well as the approval you can see the confidence score of that approval. Similarly this is the one that we just ran and we can have a look at the result and you can see that it was rejected again [music] the reasons why it was rejected and that's it. You can go ahead and upload your contracts and invoices. This application will create the indexes for you. pause all of your invoices and contracts as well as running a Llama agent workflow that handles all of the reconciliation as well as the rules that you set for that reconciliation.