Your Agreements Are a Database You Can't Query — Hiral Shah, Docusign & Sean Sodha, NVIDIA

Uh, welcome everyone to our session. Um, I'm Hiral. I'm a senior director of product, um, at Docusign, and I'm joined by Sean.

Hello, everyone. I'm a product manager at NVIDIA.

So today, Sean and I are going to talk about a massive problem that every enterprise faces, which is agreement data, large-scale agreement data. Agreements are a big part of any relationship any B2B organization kind of goes through, day in, day out, and a lot of that data is captured inside that agreements, and it's very critical, whether it's pricing tables, lot of it, and it's all in a lot of different unstructured format.

And that's kind of what we are going to show, is how Docusign partner with NVIDIA are fixing that on making that data available, readable, usable for a lot of our kind of organizations. So here is kind of just a quick map of our talk today. We'll start with just the stakes. Why does this matter? Why the scale is so large? And then we'll dive deep into the technical architecture of how we are approaching it, how we have tackled this thing, especially this document processing

at, at scale. And then finally, we'll cover what we have learned from our evaluation of all of the different models we've tried for different purposes and share our learnings with you. So why... Like, you know, just to understand why this is a big problem. When you think about Docusign... Raise your hands, how many of you have used Docusign? Anyone who's employed, probably the HR docs, right? So it's a massive scale. Everyone ha- uses Docusign. For us, it's a massive

engineering problem as well because just look at the scale. We have 1.9 million customers who are paying us and a billion users. What does that imply? We process a million agreements a day that needs to now structurize, make it readable, make it queryable, usable. And, you know, in the past, we've worked with Deloitte on a study, and it says that there's $2 trillion captured in this agreement negotiated value that no one

capitalizes, no one goes back and gets that, um, data back. Right? And why? It's because they have to do a lot of human reading, human reviews. There's disconnected systems, lot of manual workflows that are there. So that's kinda why Docusign built IAM, an intelligent agreement management platform that takes the entire, like applying in an AI-first way, the entire agreement life cycle, whether you are creating agreements, gen AI helps a lot with that, whether you're negotiating to

understanding and redlining, all the way to after signing, storing, and making a lot of insights from this data. So when you think about the challenges involved, right, in an agreement, there is the unstructured data. It could be a PDF, it could be a PNG, and what are people wanting to do is simple questions. They can't get that. That is the data that is trapped inside one agreement, but also the whole corpus of millions of agreement, ten years, twenty years

of business ha- has kind of put into that. So for a- agreements aren't flat. They're also hierarchical in nature. They are... Like, you know, one agreement governs the other, the other kind of does something else. So you always are needing lot of things to answer this question. Simple thing. I'm sure you all are using a lot of, right, Claude and all, especially at your company's organization. Simple thing. What did we contract for the total tokens, right, with Claude? No one knows. That's captured inside this agreement in different forms and fashion.

So you need to be extracting this data to find things, but also insights and push it downstream where you're tracking, doing more things. And when you analyze an enterprise contract, a big set of things are captured, what I call vital terms like pricing tiers, the SKUs, the information, SLAs, rate cards. They're all in table format. Now, traditional document extraction tools or a generic VL- VM- VLM's completely fail here. Like, you know, we've tried, we've definitely done

this. Because they're reading text line by line, which breaks a lot of that concept within the table. A merged cells, some things are not boundaried. So this makes a massive operational overhead for the downstream legal teams, procurement teams, sales teams to get that queries and get that answers done, and they spend hours and hours digging through this, um, to even just locate a basic thing. And that's kind of where we partnered with NVIDIA and leveraged a purpose-built model, like, you know, tool,

which is really es- like, you know, making our architecture for table extraction take us to make and solve these complex use cases. So we're try... With this, we are making things scalable. We can understand it with the layout, but also deliver really accurate results. And to share more about how we are leveraging the Nemotron, I'm gonna hand it to Sean.

All right. Hello, everyone. Uh, so real quick on, uh, the, uh, Nemotron Retriever initiative. So for those... Who here knows about Nemotron? Maybe raise your hand real quick. Awesome. Uh, so Nemotron is all about building world-class open source models and publishing the datasets, the techniques, uh, the quantization approaches, distillation approaches, pruning approaches, every technique possible, blueprints to go with that, you name it. Um, through then, throughout the Nemotron portfolio, we have specifically Nemotron Retriever, which is building embedding models,

re-ranking models, and document extraction models. Um, so real quick here, we-- sort of our first initiative is if you're a large-scale enterprise that deals with petabyte-scale data, our first initiative is how do we make sure that you find the right document given a certain query, your agent sends, you know, a set of queries to the corpus afterwards. Once you find those top five quer-- uh, documents, whatever it may be, then we say, "Okay, you found the right document. Now how do you then find the right information within the document?" And this is where the work with the DocuSign team has gone really great, where

we've, uh, worked with them to build the Nemotron parse model to focus specifically on table extraction, which is a really complicated technique. Um, if you think about it, the number of permutations of tables are quite vast when you think about nested tables, merged cells, merged columns, merged rows, whatever it may be, and that can get really, really complex and really hairy of a problem. Uh, so real quick, as I mentioned before, right? Our team is responsible for b-- uh, leading a lot of the leaderboards in the retrieval space. So ViDORI V1, V2, V3, MTEB, MMTEB. Um, so our

team knows how to build world-class retrieval models given a lot of leadership w- uh, given a lot of leaderboard winnings that we've had in the last year or so. And then, of course, as I mentioned before, we open source everything, right? So we share the open source model weights, the techniques, and then we release with those blueprints and skills that agents can use then afterwards. Um, so to touch a little bit on the actual model that we are working with for DocuSign was the Nemotron parse model. So when you think VLM, you generally think a multi-billion parameter model. It's very heavy. It's high

latency. Um, this is a very small, tiny C radio VLM. It's about eight hundred and fifty, nine hundred million parameter model, uh, designed to kind of be that all-in-one package sort of model where you deploy it, and instead of having small, let's say, YOLOX models that do table extraction or page element extraction or whatever it may be, this is a single-shot model that you can feed a document in and out comes the semantic formatting layouts, the text, uh, the reading order, uh, the, the preserved structure of the

table, et cetera. Um, this can be served via the NVIDIA NIM or via vLLM as well too. Um, and it's-- so it's a tiny small model that you can use. It's not a generator. It's more of an extractor at the end of the day.

Uh, so real quick as well too, um, we always wanna make sure that we're building towards benchmarks that matter most to the enterprise space. So we wanna make sure that both on the Pareto curve of accuracy versus performance, we'll make sure that we're gonna be releasing world-class models to the ecosystem too. So what you'll see here generally is just a very standard benchmark of table extraction. I believe this one was RD Table Bench. And we compare the-- some popular open source models here, and then we compare how our Nemotron parse model does compare to that industry, and we continue to kind of strive to improve this as time goes on.

So that'll-- I think we believe we have a demo as well.

Yeah.

So how does I-

You just press one... Yeah.

Okay, there we go.

Let me show you how easy it is to turn any agreement into structured, usable data with Agreement Manager, which is a central repository of every agreement an organization has ever signed. Let's look at this. So when we look at the Agreement Manager view here, you know, we have an ability to see the entire list of agreements, but also go and upload a new agreement. So I'm uploading a new order form into Agreement Manager. As you can see, I can select from a computer, import from

other places. The moment I select the agreement, it starts uploading and starts processing with AI. And just like that, you can see that the jobs engine has processed it. Let's take a closer look at this agreement. So when you go into the action, you can go and browse the file. Within seconds, Agreement Manager has extracted a rich set of metadata. Everything from key terms to commercial details are automatically structured, highlighted, and immediately you can jump

to that section where the details are found. Built-in goes deeper. This is where the NVIDIA's model comes in, that it's extracted all the structured pricing data around this agreement. It goes in, breaks down these complex tab-tables into order details. And as you can see, we can break it down. We can download all of this data. This is powered by the advanced parsing leveraging NVIDIA's Nemotron model, turning every--

even dense tables into something that is instantly usable. And of course, you can take this data with you. You can see when we've downloaded into CSV how we've structured all of it for your finance team, procurement team, even further analysis. All of it is also available through API. And that's how Agreement Manager has transformed agreements into actionable insights in seconds leveraging NVIDIA. So I think, you know, what you saw there from, um, a demo perspective, we've tried to shortened it. It's like we

have a whole repository. What you see a list, we get customers which has thousand agreements to all the way millions of agreements within. But the big piece is how do we understand and get that data that makes it very valuable to an end business user, right? A legal person, a procurement person, a salesperson who's doing a lot of the deals, or even a leader, right? Like a business unit. The CTO goes and asks, "What did we do?" This is how we are making each of the things a lot more structured. So we have our own proprietary agreement data model, which we are

structurizing each agreement, but also at a whole organization level. And leveraging a lot of the NVIDIA things, we've been able to do a really good job, especially with all of those tables, like pricing, SLAs, and then make that available. And then we also have a, like, robust kind of search that is, um, on top of it. So when you think about what have we learned, right? When you think from a Nemotron plus Docusign, we-- one of the biggest things for us, we definitely have done lot of different models for different purposes.

So purpose-built model for the job you're trying to do is a big, big part of how we've been thinking about, and that's kind of where we've been able to accelerate, bring things to market much faster. The second big piece around, like, the model efficiency. So for, you know, as Sean was talking about, the number of parameters, yes, context and stuff matters in the, you know, in a different environment for different things. For us, the lower kind of context basically also meant lower latency, lower cost to deliver the scale that we are

talking about. Last, around the faster extraction. So, um, we d- we ran this against a lot of the other open source models. When you think about how many tables can it extract per seconds, Nemotron was twenty x faster, which helps us when we're talking about the millions and billions of scale that we're kind of serving for all of our customers. So a lot of it is, like, having that smaller purpose-built things is m-- the way for an enterprise as an

organization to go and leverage and then serve that from an end user perspective. Um, and then what's next? So I'll let Sean talk through those.

Yeah. So working with the Docusign team, uh, has been awesome so far. Uh, and we're gonna continue to deepen that partnership as well over the next few months. So, uh, with them, we started with the, how do I extract as much possible information from a page, and now we'll scale to how do I now find that page to begin with. Um, so we'll start a little bit with the Nemotron, NeMo Retriever, uh, effort, and then, of course, we'll talk a little bit about the NVIDIA Agent Toolkit with them over the next few months, um, and then actually start scaling into, into more production scale agents then.

Perfect. I think that's what we had. We have time for a couple questions, anyone in the room.

Okay. Have someone there.

It seems like the, uh, OCR is, uh, a bottle already decided when you guys were thinking of basically cut that out of the loop and building an agentic structure from the get-go moving forward so that basically only the core relevant structure remains.

So just to recap for everybody, if you didn't hear, it was-- the question is, right, like, OCR is always a thorn in the whole process, so are we thinking about letting the go of that and starting from agentic from the get-go? I can talk from my perspective. So I think for us, right, like, there are different use cases at different points in time. Many times, if you are reactive, you have a question and you're coming, some of that can, uh, can, can work dynamically at a smaller scale. The question is the latency. When I am querying at that scale of thousands, I do

need to have preprocessed, have identified, so that's one. I think the second big part of the use case for us, a lot of times businesses want to use this data to do a lot of downstream work. So an example is a procurement team. This is my pricing table. I wanna put it into Coupa to make sure when I'm paying that works. At that time, there i-- like, you know, the agent is kind of helping, but I can't do that on a one document by document. That said, there is ways that we are compressing. That's kind of why Nemotron worked for us, is like, how do you do it from a layout understanding just for that

purpose? But I would let, let you add.

Yeah. I think it depends on the use case a little bit. Um, I think for this specific instance, right, you have petabytes of documents that you want to be queryable at some point, right? So you are heavy on the compute at the upfront side with all the OCR, so you don't have to worry about it later on, right? Then there's some instances where people may upload a contract to begin with for Q&A, and that's a very high-- that's a very low latency use case, right? So you have a high throughput versus low latency use case. And in that scenario, your different batch sizes, your concurrencies, your different techniques on how you process the document will be different. And where you

spend that compute in that cycle will be changing between the different use cases.

Okay. One more there. Yeah.

I have a similar question. So next, how this solution replace YOLOX OCR tool since it's able to just extract new document. How, uh-

We, we do a lot of, like, more of what I call hybrid approach at a purpose-built for, like, the needs and the use cases. So from a table piece, it does kind of, you know, do the whole layout along with extracting. We still do OCR from a lot of other fields and metadata in the clauses, like all of the text kind of thing. So we-- the-- I think we had a architecture where we have a pipeline going through two different routes for that. Um, as a follow-up, we have a blog out there. How are we really solving this at scale across? And if you look at that, there's a lot of different piecemeal

modules and stuff together. Yeah, one more. At last.

Uh, at scale, what quantization techniques are you using with the Nemotron model?

So we-- this model is currently on FP16, but there are paths towards going on to FP8 and then FP4 in the next few months as well too.

What about the Docusign? Are you, like, using the sixteen or are you quantizing down?

We do use that, and then we are also kind of using some of the older ones, and that's the journey as a partnership, is to kind of go tweak as you get more of the customers in.

Yeah. So for this, there are many techniques on how to improve the performance side, right? So quantization, right? So we're trying to move everyone to Blackwell, right? So that's why NVFP4 is the big thing now. Um, as well as, uh, multi-token generation for this. It's a VLM architecture, right? So your encoder-decoder techniques can definitely be further optimized. So not right now, this model just generates one token at a time. You can do multi-token generation, of course, too. So there's plenty of performance things. Right now we're focusing on the accuracy side. Like, are we adding value to the system? And then from there, we'll then push out that Pareto curve on the performance side.

Are you guys using, like, the, the category graphs on NVIDIA GPUs, or are you, like, primarily running through force?

I believe they just deploy via vLLM directly.

No, I've got one on the recall side. When you, when I actually... Like, when you did that demo and you asked the question, is that, like, is, is that graph to the embedding structure, is it using camera or are you just using brute force?

Oh, th-this is just an extraction. This is not a retrieval.

Yeah.

Yeah.

Oh.

Yeah.

Even when you're asking the question, was it a retrieval?

It's coming from that agreement data that we've kind of extracted and stored.

Yeah, maybe I can chat with you offline and how, like, we, our architecture kind of works fully as well.

No, we're almost coming up on time there. Um, but I think that's kind of all we have. Um, happy to hang around, uh, in the back with more questions. Um, and good luck with a lot of your, uh, challenges with AI. So thank you.

Thank you.