Video transcript:
Good day and welcome. My name is Benedikt Hausberger. I'm a software developer at CIIT and have recently specialized in machine learning projects. Today, I'd like to introduce you to a prototype we designed and developed in collaboration with Dlouhy GmbH Fahrzeugtechnik. This prototype simplifies the accounting of vehicle costs, making it easier to record and save expenses. It's integrated into the existing and operational ObJack application. ObJack is a management system for objects of all kinds. In this specific case, we have vehicles and vehicle-related objects. You can then store information under "objects," such as what's inside the vehicle. You can create jobs, reports, or damage reports, run routine tasks, add data or resources, and, of course, record costs. An example of a cost entry or an invoice is to enter it. This is a sample invoice with several line items. Among other things, it might include: diesel fuel was added to the vehicle, diesel fuel was used, maintenance was performed, the engine oil was changed, and the Aral Card was billed online. This is a typical monthly statement from Aral AG. This one is for November, where all the items are saved as an image or PDF. It shows both net and gross amounts. You can enter this sample invoice manually, meaning you get the costs included. You have to enter the invoice number, the supplier, the invoice date, and all the items. You can also enter a cost type, for example, fuel. Then you have to enter the net costs. You need to describe what was fueled, and you have to do this for all the items I showed earlier. It's important to note that these are only the net costs. You have to calculate the taxes separately. What we've developed is a prototype, or a service, that automates exactly this process. And I'll show you how it works. First, take this one and test it. You upload it, and then you'll see the button "Extract Items Using AI." That's the image. Then you send it to our service, which then sends this image to the Azure Cloud, loads it into, for example, the model, and then omits all the line items, and we insert it into our application. Right, then you can see immediately, okay, it's happening right now, the invoice is being extracted. You also see an indicator that it's happening in the background, and you don't have to do anything. Just wait a little. This usually takes 5 to 15 seconds, depending on the size of the invoice, and then you can see all the data and all the line items. What's also very important to us is that you can always see what has been generated, and you can see that right here, generated and also highlighted in red. Why red? The color indicates that it hasn't been signed yet and needs to be verified. Because you can never be 100% certain that what the AI generates is correct. You can never be sure. And especially with this data, where it's about invoices and taxes, it simply has to be accurate. This means we've implemented a system where the invoice is only accepted once it's been signed and verified by a user. When you access the invoice, you'll immediately see that it needs to be signed and confirmed. You do this by clicking on the "sign" button. Then you'll have all the necessary information: the invoice date, the supplier or provider (i.e., who issued the invoice), and the invoice numbers. You'll also see all the individual items, including their type (e.g., "fuel"), a description of each item, and finally, the tax. If you notice a small error, like a few cents off, you can correct it. This system works quite well. You can also add further information to the description if you wish. To do this, you simply need to verify and sign the invoice. And when you click on it, it gets signed in our application, and then you can see here, okay, it's still generating wars, but it's already signed. This process is significantly faster than manual entry. You don't have to do that; it's generated or extracted by the AI itself. So, the big advantage of this application is that it's simply much faster and much easier. This means you can save a lot of accounting work that's normally done by two, three, or four people. Well, that concludes the presentation. As we've seen, this can save a lot of work very quickly, without you having to do anything yourself. Simply capture the invoices, recognize them, and insert them into the main application. And this service can then be used in many other areas to extract similar information from PDFs or images and process it further in applications. Thank you for listening, and see you next time.