Hyperscience Platform Demo

What you’re looking at here is the Hyperscience UI. Hyperscience is a SaaS solution, though we do also offer an on-prem solution. Regardless of which deployment option you choose, users will access the application through a web interface, which is what we’re looking at here. Today I’m going to submit a PDF file into Hyperscience. Remember Hyperscience ingests images, and we’ll see how Hyperscience is able to classify those documents, find and extract the data from those documents, intelligently looping in a human where required, and then provide a structured data output, which would then flow downstream into a system of record and ERP system.

I’m dragging a PDF file which I have stored locally on my computer and I manually submitted that into Hyperscience using the create submission button. Keep in mind that is purely for demo purposes. As I mentioned when we were going over the overview, we have a library of connectors, things like email listeners, RPA bots we can integrate with, which makes it easy to snap this into whatever your document processing workflow or tech stack looks like. So in production, you’re not going to be manually submitting images or files into Hyperscience; all of that will be programmatic.

While we give this a minute to process, I’m going to open up this file and show you what I’ve actually submitted. This is a three-page PDF file. Our first page is a W-9 form; you’ll notice it’s upside down and has also been filled out by hand. Our second form is another W-9, mostly computer printed but it does have some handwritten corrections on the form, and the image itself looks like it got skewed up or twisted up in the scanner or in the fax machine. Finally, I have a Social Security Administration form. This is also handwritten in a fairly low quality document. It’s very noisy, looks like there’s some fax header information here, and it looks like the document has been shrunk down to maybe 90 or 85% of its original size.

I’m going to close this file and click into our submission. Here I’m able to see that Hyperscience was able to separate and classify each one of those three documents. So we have our IRS W-9, we had two of those, and then our Social Security Administration form. If I take a look at the status of these documents, you’ll notice that these are in a manual transcription status.

I’m going to tell you what this means exactly. One of our big differentiators in the market is that you are able to set your own accuracy within the Hyperscience Platform. Many of our customers require 99% accuracy or above. We measure accuracy as well as automation at the field level. So what that means is let’s say you need 99% accuracy; there is an error budget of one field in a subset of a hundred fields on average. We think about automation in the same way. So 80% automation translates to, in a subset of a hundred fields, 80 are being automated by the machine and 20 are being manually handled. We run our demos at 99%, and where our machine is not confident that it can deliver the accuracy SLA, in this case 99%, it’s going to raise its hand and ask a human for help. That’s exactly what you see here.

I’m gonna go ahead and click into this UI now. Here I’m taken to the first document and Hyperscience is asking me for help on this name. I’m going to key in what I see and I think this looks like Thomas Edison, so I’ll simply key that in, click enter on my keyboard. And that was the only field on this entire document where Hyperscience needed help. I’ll click enter again on my keyboard to submit that document. Now I’m taken to the next field on the next document and you can see here this is an account number. Again, I’m going to go ahead and key in what I see.

Just a quick note on this UI: it has been designed for high-performance data keying. Everything can be navigated on the keyboard. There is no need for keyers or processors to pick up their hand and move back and forth between the mouse and the keyboard, and we actually have the keyboard shortcuts listed right here in the UI. So here’s our account number. I’ll click enter to submit. And again, that was the only field on this document where Hyperscience needed help. I’ll click enter again.

We’ll let Hyperscience finish up processing these three documents and then we’ll take a look at some extractions. I can now see that each one of these documents has been completed. You’ll notice my SSA form was a hundred percent automatically transcribed by the machine. So that document was straight-through processed. It needed no human touch whatsoever. And then of course our two W-9s I had to help out the machine on one field, but the vast majority of the fields were automated by the Hyperscience machine.

One way to think about this is if you can achieve 80% automation, which is generally for structured forms we see around 70 to 80% automation on day one, as our machine gets smarter that will increase all the way up to sometimes 90-95% automation. It’s completely dependent on your forms and documents. But think about that is a significant reduction in average handling time. If it takes a keyer 10 minutes to key in this document today and you can reduce that by 90%, it will only take one minute to support the same amount of work.

Let’s take a look at some of these documents. So here is this Thomas Edison name field and you’ll notice that I’m able to see that I was the one that transcribed this field. We have comprehensive reporting available as part of the platform. I’m able to see how much of the work is automated, how much is performed by humans as well as the accuracy of both the Hyperscience machine and the accuracy of individual keyers who are handling these documents. I’ll compare that to General Electric, the business name, and I can see that the machine transcribed this field here.

Here’s a great example of where Hyperscience understands human intent when looking at these forms. I’m going to zoom in a little bit. Here’s the social security number and you’ll notice that the number nine is falling outside of the space on the form for the social security number and outside of the bounding box where we’ve asked Hyperscience to look for the data. Now what you can see though is that Hyperscience is still able to capture the entire social security number and that’s because it knows what a social security number looks like, it knows how many digits it has and it knows that it can look a little bit outside of this blue bounding box for that data.

Let’s take a look at our second W-9 here. Another great example of Hyperscience being able to understand human intent. Here in the business name, whoever filled out this form made a mistake and you’ll see that Hyperscience is smart enough to ignore the cross-off. We can also read checkboxes; ‘true’ would indicate checked, ‘false’ would indicate unchecked.

Finally, here is our SSA form. This was a really messy form. I’ll take a look at the date. It’s June 12, 1987; you’ll notice there’s some normalization being applied to the raw data. You can completely customize those normalization rules. And another great example of I think a really impressive extraction where other technologies in the marketplace specifically OCR would really struggle is this street address. If I just take a look at this field here, me as a human I can guess that this probably says 111 Illinois Avenue, but the characters on the page are really just a bunch of vertical lines. I’m not sure what’s a ‘1’, what’s an ‘i’, what’s an ‘l’. Because Hyperscience has a notion of what an address looks like because we defined a data type for every single field, it is able to accurately discern the 1s and the Ls and the Is and ultimately accurately extract this field of data.

Just to bring this full circle, so at this point our documents have been classified, the data has been extracted at a very high degree of automation and then here’s an example of what that structured output could look like which would then flow downstream into a system of record or an ERP system.