AIP Podcast EP 63 - Structure Document Data at Point of Capture with Discrepancy AI

This episode's guest, Lisen Kaci, Co-founder and CEO of Discrepancy AI, discusses how his team's solution aims to become the Stripe for no-code document processing. Built to index and analyze complex unstructured documents while detecting anomalies, fraud, and security issues, Discrepancy AI works with Images, unformatted PDFs, forms, charts, tables, financial records, tax documents, receipts, invoices, bills, and a picture of a scan photocopied eight times.

Follow AIP Affiliate, Discrepancy AI
Website: https://www.discrepancyai.com/
LinkedIn: https://www.linkedin.com/company/discrepancy-ai-company/

Follow AI Partnerships
Website: https://www.aipartnershipscorp.com/
LinkedIn: https://www.linkedin.com/company/aipartnershipscorp/
X: https://twitter.com/AIPartnerships

The AIP Podcast is hosted by Anne Cheng, on behalf of the AI Partnerships, a Railtown company

Transcript

SPEAKER_01 0:05

Imagine a short code that plugs into any website or platform that instantly structures all your data, tables, images, and text at the point of capture. And here's the kicker. Not only is it simple to get started, it's cost-efficient to keep going. On today's episode of the AIP podcast, we're going to learn about discrepancy AI's mission to get rid of a problem that over 82% of all organizations globally still face. Document Wrangling. Hey everybody, welcome back to yet another episode of the AIP Podcast. And on behalf of the AI Partnerships Core, it's Ann Cheng, your host from Supercharged Lab. And boy, do we have a fun episode for you. Today, we're tuning in to learn more about Lyson Casey. He is the founder and CEO of Discrepancy AI, an up-and-coming AI startup out of Toronto that was part of Techstar's Washington, DC 2024 and is just about to blow document wrangling out of the water. No more pulling your hair out to try to figure out contextual information with millions of images and badly scanned documents. Discrepancy AI's mission is to index and analyze complex, unstructured documents while detecting anomalies, fraud, and security issues. Let's jump right in. Langza, super good to have you here with us today. Tell us about your story. How did you get started?

SPEAKER_00 1:36

Yeah, thanks for having me, Ann. This is uh really great. Uh I really appreciate you taking the time and I'm super excited to be here. Uh well, my start is I'm an AI engineer by trade, and I was um I was a lead engineer uh at a company and I had to work with uh medical documents to uh to review them, and that's really where I saw the issues with the current state of the art called OCR, and uh also how complex it was to work with your with user documents and to kind of build processing pipelines to to work with those. Um so that's where I decided there needed to be a better uh a better solution. And uh after working with that, I built discrepancy AI, which uh converts any document directly into the structured data or technically JSON, uh we call it, that your company needs. And uh above that, after working with so many customers, we realized that hey, they don't have all of the software engineers in place to be able to handle the front end of capturing these documents, the database, the servers, the AI integration. So that's why we created this vertical slice to really handle document processing and capturing for our customers. And we always say we want to become the stripe for document processing. So, kind of what they did for credit card processing by allowing businesses uh to really easily capture these document uh these credit cards, we want to do the same thing for documents. Have a no-code, low-code solution that allows businesses to capture documents and process them all without needing complex coding uh or uh engineering in place.

SPEAKER_01 3:14

That's amazing. You know, traditional approaches to contract management AI and document handling AI have typically been with the use of OCR, you're right, or even optical character recognition. Now, discrepancy AI is completely different in that it does not use OCR at all. Can you tell us a little bit more about that?

SPEAKER_00 3:35

Yeah, so we use uh a series, uh we call it AI tool chaining. So we use a series uh uh of uh multi-channel LLM, omni-channel LLMs to really work, uh analyze the documents like a human would. Um so OCR, if you don't know, just converts the extracts the text from these documents and gives you like an XY coordinate in the table for where that text is. Then uh as a developer, which is what I used to do, you have to process these texts and extract insights and do all these things from them. And it's very difficult to know what that text represents. So that's why uh a discrepancy AI, we I knew that the solution, since everybody uh wants to end up with JSON, why not have the AI look at the document and create the JSON without any middle steps? So that's why the the the we I decided to go with this direction uh of having these uh cutting-edge AI models uh instead of relying on OCR, which was uh created in the 70s, believe it or not.

SPEAKER_01 4:37

Well, that's that's impressive. I did not know that. Um, but you know, during our pre-call um conference, as well as just now, you mentioned something that's really revolutionary, and I don't know if our audience picked it up. Discrepancy AI does the structuring of data at the point of capture, but with different customer needs and data schemas. How do you create a SaaS solution without significant customization required?

SPEAKER_00 5:04

Yeah, so uh the answer is that there is significant customization, and we've built a system that allows users to uh customize their system using just natural language by defining what they want extracted on our platform. And then they can just paste a link on their website or app to allow their users to upload these images, upload their documents, and um get the insights uh right at the front end, right as the user uploads these documents. So that way, structuring at the point of capture, we can provide feedback as well as make sure that the correct documents are uploaded. And uh our system allows the customizability uh without having to code and uh you're really defining the extraction criteria like you would a person uh that is uh working to extract and capture these documents from you.

SPEAKER_01 5:55

That's that's so critical. You know, during our pre-show call, you also mentioned that you started out in the legal document space, which is so fragmented, but then you pivoted. How did you figure out the need to make that pivot?

SPEAKER_00 6:10

Yeah, so I I I spoke to hundreds of lawyers and legal professionals and legal ops, um, and I learned something pretty critical, uh, a little controversial, but uh I find I'll say it that uh if certain legal firms that are um charging via billable hours, uh AI is a threat to them, and they uh do not like uh a product or a system that will decrease their billable hours. So there's still certain legal professions where AI will be incredibly helpful and beneficial. So in-house, for example, uh in-house counsel, they're not charged, they're not, they don't have billable hours. So if you can uh increase the amount of work that they do, that would be great. So um it's not that it's not great for all lawyers, but you have to be specific in targeting them. Uh at discrepancy, I realized that there was a lot uh lower hanging fruit where we could solve a much more pressing challenge that current OCR technology wasn't. And that's really with financial documents, uh charts, tables, forms, handwritten documents. These are um these are the types of files that current technologies fail at uh and where we can provide the most impact for our customers.

SPEAKER_01 7:22

Yeah, that's that's critical. Um, and it's a really controversial view about the legal for uh ecosystem. But you know, onto this whole you know proliferation of AI startups, there are a lot of them in the same space, and they're all trying to tackle the same challenge of finding a way to structure unstructured data. So, what sets you apart?

SPEAKER_00 7:48

Yeah, that's it's an incredibly crowded field, but it is not a solved problem. That's what's really interesting about this. 90% of data is still unstructured. Um, and what makes us different from every other uh company trying to structure uh documents is that we are targeting non-technical folks, uh service companies, uh product leaders. Our solution is not targeted at developers or technical people. Every other solution out there is. It requires a developer to set up, um, to handle and to process, whereas our solution is fully no code to low code. Uh it's tailored at service professionals that don't have uh a dev team or product leaders that want to really easily build customer-facing products that capture documents from their users. So that's really who we who we are tailored to. Um medium to large enterprises that um don't aren't tech companies per se, but they want to process these documents from their users without needing to put together uh a six to eight person uh tech team, dev team, to be able to build these entire document processing workflows.

SPEAKER_01 8:57

That's uh so critical um really to sort sort of even dumb it down. Um, but on to the horizon for you and discrepancy AI, really, now that you're done with tech stas, what's what's on the horizon for you and what are you really heads down working on?

SPEAKER_00 9:17

Yeah, so so right now um we're really heads down at uh uh gaining as many customers and solving as many problems for people as we can. Uh so that's our main focus is getting out there, uh helping uh helping as many service industries um and being really the go-to place for for any uh business that wants to capture documents from their users. Uh I always say we we are like the stripe for document processing. So we want to be uh in as many document uh workflows as we can and help as many companies as we can uh understand the documents that are coming into their system.

SPEAKER_01 9:57

That's great. Lizen, listen, it's been a blast, and I'm sure you had a lot of fun as much as I did. So, to all our listeners and all our subscribers on Spotify, YouTube, and LinkedIn, your support has been incredibly valuable to us. So please don't forget to like, share, and follow us. It will help us as much as you'll you'll never know, really. Uh, once again, my name is Anne. I am your host on behalf of the AI Partnerships Core, and I've been speaking to Lizen Casey of Discrepancy AI. Thank you for sharing your time with us.

SPEAKER_00 10:33

Thank you so much for for having me. It's been a blast, and yeah, I I loved uh speaking with you and telling uh you and your viewers a little bit about uh discrepancy AI.