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

Hiral Shah, Docusign · Sean Sodha, NVIDIA

[Talk page](https://ai.engineer/talks/_gvamfT8H-w-your-agreements-are-database-you-cant-query) · [Video](https://www.youtube.com/watch?v=_gvamfT8H-w)

## Your Agreements Are a Database You Can't Query

Hiral Shah and Sean Sodha explain how a small, layout-aware model turns contract tables into structured data, why large repositories need preprocessing, and where OCR and accuracy work still matter.

![Selected presentation frame from Your Agreements Are a Database You Can't Query — Hiral Shah, Docusign & Sean Sodha, NVIDIA at 589 seconds](https://ai.engineer/api/talk-essay-media/88c5ca3c2622fa443ee3d6d282eade6f5782a31252927e70d43fd858d1184d3c)

Making Agreement Data Queryable: Docusign’s Purpose-Built Approach to Table Extraction

### The signed agreement is only the beginning

[0:13](https://www.youtube.com/watch?v=_gvamfT8H-w&t=13s)

A company can negotiate pricing carefully and still struggle to recover what it agreed to buy. [Hiral Shah](https://www.docusign.com/blog/author/hiral-shah), a senior director of product at Docusign, opens with this everyday problem: important business terms sit inside signed agreements in forms that other systems cannot readily use. With [Sean Sodha](https://www.linkedin.com/in/sean-sodha), a product manager at NVIDIA, she explains how Docusign is turning that document collection into structured, queryable information.

The workload is substantial. Shah reports 1.9 million paying customers, a billion users and a million agreements processed each day. She also cites a study conducted with Deloitte estimating $2 trillion in negotiated agreement value that organizations do not capitalize on. Recovering the terms requires people to read documents, review them and move information through disconnected systems. The estimate describes unused negotiated value; it does not measure savings delivered by the extraction system.

Docusign’s intelligent agreement management platform, IAM, spans creation, negotiation, redlining, signing, storage and analysis. After signing, the challenge becomes answering questions from PDFs, images and repositories containing ten or twenty years of business history. The documents also have a hierarchy: one agreement can govern another. A useful answer may therefore require information from several related agreements.

Shah makes the problem concrete with a company asking how many total tokens it contracted for with Claude. The terms exist somewhere in the agreements, expressed in different forms. Finding the relevant words is only part of the work. The system must recover their meaning and organize them so the company can ask questions across its agreements.

![Selected presentation frame from Your Agreements Are a Database You Can't Query — Hiral Shah, Docusign & Sean Sodha, NVIDIA at 212 seconds](https://ai.engineer/api/talk-essay-media/c56853e57537175b6b87280f725750f2083137ec00f64e21f6aae6044fcb7be8)

The signed agreement is only the beginning

### Tables carry relationships that plain text loses

[3:52](https://www.youtube.com/watch?v=_gvamfT8H-w&t=232s)

Pricing tiers, SKUs, service-level agreements and rate cards often appear in tables. Shah says traditional extraction tools and generic vision language models failed on the tables Docusign tested. The failure is structural: reading a page line by line can retain the text while losing which values belong together.

A merged cell can apply across several rows or columns. Once a parser flattens the table into a sequence of words, that scope becomes harder to recover. The extractor needs to preserve the arrangement of the table as well as its contents, because the arrangement helps determine how the commercial terms apply.

![Selected presentation frame from Your Agreements Are a Database You Can't Query — Hiral Shah, Docusign & Sean Sodha, NVIDIA at 292 seconds](https://ai.engineer/api/talk-essay-media/4bfde68ae53650c2ac030a41b35b9a3aa11b19a77a797ddc638a17fb9072f29b)

Tables carry terms that plain text loses

When the structure disappears, legal, procurement and sales teams must reconstruct it from the original page. Shah describes people spending hours even locating basic terms. Docusign’s partnership with NVIDIA targets that particular source of manual work: table extraction that understands layout, so downstream users receive structured commercial data.

### Find the document, then recover its structure

[5:17](https://www.youtube.com/watch?v=_gvamfT8H-w&t=317s)

Sodha places extraction within **Nemotron Retriever**, NVIDIA’s work on embedding, re-ranking and document-extraction models. He separates two tasks: finding the right document in a petabyte-scale corpus, and finding the right information inside that document. Returning a handful of relevant documents still leaves the system with the job of reading their tables correctly. Docusign’s collaboration focuses on that second task.

Sodha expands on why tables are difficult: nested tables, merged cells, merged columns and merged rows create many possible layouts. NVIDIA’s surrounding approach includes publishing model weights, datasets, techniques and supporting blueprints. He names quantization, distillation and pruning among the techniques shared with developers. The model at the center of this integration is **Nemotron Parse**, which recovers document structure and is used here for table extraction.

### What one model invocation returns

Sodha describes the parser as approximately 850–900 million parameters, much smaller than the multi-billion-parameter vision language models he contrasts it with. A single invocation consolidates work that might otherwise use separate models for tables and page elements. Its outputs preserve several distinct parts of the document:

- **Text and semantic formatting:** the content and its organization.
- **Layout and reading order:** where elements belong and the sequence in which they should be read.
- **Table structure:** the relationships needed to keep complex tables intact.

The model can be served through NVIDIA NIM or `vLLM`.

![Selected presentation frame from Your Agreements Are a Database You Can't Query — Hiral Shah, Docusign & Sean Sodha, NVIDIA at 464 seconds](https://ai.engineer/api/talk-essay-media/4921930db2dc3e67e27a0d380ce6f555f65d41a47385ff1c21431c26b3ed4fbf)

Finding the document and reading its tables are separate jobs

Sodha calls it an extractor because the task is to recover the supplied document’s content and structure. This describes its purpose. As he explains later, the model still produces its output one token at a time. A single document-processing invocation can therefore contain sequential token generation; consolidating the extraction tasks does not make decoding simultaneous.

NVIDIA aims to improve the accuracy-versus-performance Pareto curve: give enterprises a better combination of extraction quality and speed. Sodha introduces a table-extraction benchmark comparison against other open-source models. His spoken explanation does not provide numerical accuracy scores or enough comparison detail to reconstruct a ranking.

### An order form becomes usable agreement data

[8:30](https://www.youtube.com/watch?v=_gvamfT8H-w&t=510s)

The Agreement Manager demonstration follows one order form through the product. Agreement Manager is a central repository of signed agreements, with options to upload a file from a computer or import it from elsewhere. Selecting the order form starts uploading and AI processing; a jobs engine handles the processing before Shah opens the file to inspect the results.

### Inspect a term or take the data elsewhere

The walkthrough presents two useful ways to work with the extraction:

- **Return to the source:** key terms and commercial details appear as structured metadata, with highlights and navigation to the section containing each detail. A reader can inspect a term in its original context.
- **Use structured pricing:** NVIDIA’s parser breaks complex pricing tables into order details. Shah downloads the data as CSV for finance, procurement or further analysis, and describes API access as another way to carry it into other systems.

![Selected presentation frame from Your Agreements Are a Database You Can't Query — Hiral Shah, Docusign & Sean Sodha, NVIDIA at 574 seconds](https://ai.engineer/api/talk-essay-media/3515871facd5349bbadda732d59916a4a8210215c0f9e70e989e40e084fdc963)

An order form becomes data people can use

Shah describes results appearing within seconds, then notes that the demonstration was shortened. It illustrates the upload-to-output flow without establishing a general latency guarantee. [Docusign’s companion integration post](https://www.docusign.com/blog/how-docusign-is-bringing-contract-table-extraction-to-production-with-nvidia-nemotron-parse) describes Nemotron Parse table extraction in Agreement Manager as accepting beta customers, with general availability ahead. That is the availability stated in the post, rather than a determination of availability at the recording date or today.

API availability also needs that distinction. Shah describes API access in the walkthrough, while the companion post describes a public API for direct downstream integration as forthcoming. The demonstrated export flow and the planned public integration interface should therefore be understood separately; the descriptions do not establish general availability of a public API.

Customers’ repositories range from thousands to millions of agreements. Docusign uses a proprietary agreement data model to structure information both within each agreement and across an organization, with search above that model. This lets the extraction serve legal staff, procurement teams, salespeople and business leaders. Shah does not explain the schema or how the system resolves the governing relationships between agreements, so the parser’s table output provides only part of the explanation for organization-wide answers.

### Specialization changes the economics of extraction

[11:36](https://www.youtube.com/watch?v=_gvamfT8H-w&t=696s)

After trying different models for different purposes, Shah’s central decision is to choose a model for the job it needs to perform. She credits purpose-built models with helping Docusign bring capabilities to market faster. In this workload, using less context also reduced latency and delivery cost. Parameter count and context size are separate considerations, but both matter when planning extraction at a million agreements a day.

Shah reports that Nemotron was twenty times faster than the other open-source models Docusign tested, measured in **tables extracted per second**. That throughput gain matters for a large ingestion workload. It does not establish a twentyfold improvement in the entire agreement pipeline. She does not specify the comparison models, hardware, batch sizes or absolute throughput, so the result remains tied to Docusign’s reported evaluation.

![Selected presentation frame from Your Agreements Are a Database You Can't Query — Hiral Shah, Docusign & Sean Sodha, NVIDIA at 736 seconds](https://ai.engineer/api/talk-essay-media/3f6dc94f5d2c2b46f849540e14ca6bd6ae19b72aa4e6a18adc85077d8ada6db3)

Specialization changes the cost of extraction

The partnership’s next planned step moves from extracting information on a page to finding the relevant page in the first place. Sodha describes further retrieval work and discussions around NVIDIA Agent Toolkit, followed by more production-scale agents. These are proposed extensions of the collaboration, rather than features demonstrated in the order-form walkthrough.

### When to spend compute—and why OCR remains

[13:12](https://www.youtube.com/watch?v=_gvamfT8H-w&t=792s)

The Q&A asks whether an agent could bypass optical character recognition, or OCR, and extract only the structure relevant to a question. Shah allows that dynamic processing can work for smaller, reactive use cases. When a query spans thousands of agreements, however, she wants the documents already processed and the relevant information already identified. Otherwise, document processing adds to the user’s wait for an answer.

Downstream operations provide another reason to extract terms ahead of time. Shah’s example is a procurement team putting a contract’s pricing table into Coupa to help payments match the agreed prices. That workflow needs structured terms available for continuing business use. An agent may assist, but answering questions one document at a time does not meet the whole requirement.

### Repository throughput and immediate Q&A need different settings

Sodha explains the choice in terms of when the system spends compute:

- **Large repository:** process the collection upfront, including OCR, so later queries can use information already extracted. The ingestion workload emphasizes throughput across many documents.
- **New upload for Q&A:** process a newly submitted contract quickly enough to support an immediate interaction. The user’s wait makes latency more important.

Batch sizes, concurrency and processing techniques should change with those goals. The engineering decision includes the processing schedule as well as the model.

A follow-up question clarifies what Docusign retains: a hybrid pipeline with two processing routes. The table route combines layout understanding and extraction. OCR still handles other fields, metadata and clause text. The small parser consolidates work for tables, while other document-processing modules remain necessary. The flow below shows those routes and the later uses of extracted data, without specifying an internal routing or search algorithm.

Two extraction routes serve stored agreement data
The table parser preserves layout and structure; OCR handles other content. Extracted data supports later questions and downstream work.
- Agreement document — Contains tables, fields, metadata and clause text.
- Table route — Layout-aware parsing extracts structured table data.
- OCR route — Extracts other fields, metadata and clause text.
- Stored agreement data — Extracted information is organized for later use.
- Questions and search — Use agreement data already extracted and stored.
- Downstream work — Pricing data supports procurement and payment workflows.
- Agreement document → Table route (Tables)
- Agreement document → OCR route (Other content)
- Table route → Stored agreement data (Structured terms)
- OCR route → Stored agreement data (Extracted fields and text)
- Stored agreement data → Questions and search (Later use)
- Stored agreement data → Downstream work (Export or integration)

### Accuracy first, then precision and decoding optimizations

[16:09](https://www.youtube.com/watch?v=_gvamfT8H-w&t=969s)

Asked about quantization, Sodha says the model currently uses FP16, with paths toward FP8 and FP4 over the following months. These are numerical precision formats: the numbers refer to 16-, 8- and 4-bit representations. Moving toward lower precision is one proposed way to improve inference performance. He names Blackwell and NVFP4 as part of NVIDIA’s direction. Shah indicates that Docusign uses the discussed precision and some older models, but does not give a precise breakdown of its deployed configurations. No measured production result from the proposed lower-precision changes is reported.

### The remaining performance levers

Sodha identifies several ways to improve the parser:

- **Quantization:** move toward lower-precision formats as part of the hardware and inference strategy.
- **Multi-token generation:** improve on the current one-token-at-a-time output process by generating multiple tokens at a time.
- **Encoder-decoder optimization:** further tune the VLM architecture’s encoding and decoding.

These are possibilities, rather than reported speedups. His priority is to establish that the extraction is accurate and adds value, then improve the performance side of the Pareto curve.

The final questions return to deployment and search. Sodha says he believes Docusign deploys directly through `vLLM`. [Docusign’s companion post](https://www.docusign.com/blog/how-docusign-is-bringing-contract-table-extraction-to-production-with-nvidia-nemotron-parse) explicitly confirms that serving choice and says sensitive agreement data stays within Docusign’s environment. The post supplies that deployment detail, while Sodha’s answer itself remains tentative.

An attendee then asks about the retrieval mechanism behind the demonstration. Shah’s answer is narrower: the information comes from agreement data already extracted and stored. She does not specify an embedding index, graph-based retrieval method or search algorithm. The explanation ends with a useful separation of responsibilities: extraction makes the contract terms available as data, and the system’s question-answering features consume that stored information. How those features select the relevant data remains unspecified.

## Key takeaways

- Useful table extraction preserves cell relationships, merged structure and reading order. Recovering the words alone can lose the meaning of commercial terms.
- The approximately 850–900 million-parameter Nemotron Parse consolidates several extraction tasks into one invocation. Docusign still uses OCR for other fields, metadata and clause text.
- Docusign reports twentyfold higher table-extraction throughput against the open-source models it tested. The result does not establish the same improvement across the complete agreement pipeline.
- Preprocessing makes large repositories ready for later questions and downstream workflows. Immediate Q&A on a new contract has different latency, batching and concurrency needs.
- FP8, FP4 and multi-token generation are proposed performance improvements. Sodha prioritizes accurate, useful extraction before further speed optimization.
- Shah says answers use agreement data already extracted and stored. The closing exchange does not explain the search algorithm that selects that data.

## Resources

- [How Docusign is Bringing Contract Table Extraction to Production with NVIDIA Nemotron Parse](https://www.docusign.com/blog/how-docusign-is-bringing-contract-table-extraction-to-production-with-nvidia-nemotron-parse) — Companion explanation of the layout and OCR integration, vLLM serving, data handling and evaluation on real enterprise contracts. The supplied post describes table extraction as accepting beta customers, with general availability and a public downstream-integration API ahead.
- [How to Build a Document Processing Pipeline for RAG with Nemotron](https://developer.nvidia.com/blog/?p=112323) — A separate NVIDIA tutorial explaining how extraction outputs feed embedding, retrieval, reranking and cited answer generation. Its pipeline does not establish Docusign’s implementation.
- [Build an Enterprise-Scale Multimodal PDF Data Extraction Pipeline with an NVIDIA AI Blueprint](https://developer.nvidia.com/blog/?p=87948) — An earlier modular extraction architecture using separate detection, OCR and retrieval components, useful context for the tasks the small parser consolidates.

## Slides

### Your Agreements Are a Database You Can't Query

Hiral Shah and Sean Sodha introduce Docusign and NVIDIA’s work to make unstructured agreement data available, readable and usable for organizations.

[0:36](https://www.youtube.com/watch?v=_gvamfT8H-w&t=36s) [Title slide introducing Docusign and NVIDIA presenters and their work on agreement data.](https://assets.ai.engineer/sites/aiecode2025/sha256/69763faf31a67f3c80cc9fdc3a651a1382833442306a8f66aa870514e14e759a.webp)

Your Agreements Are a Database You Can't Query.

We're Fixing That.

Hiral Shah
Product
docusign

Sean Sodha
Product
NVIDIA

### What We’ll Cover Today

The talk moves from why agreement data matters and is hard to understand to Docusign and NVIDIA’s document-processing approach and lessons from model evaluations.

[1:05](https://www.youtube.com/watch?v=_gvamfT8H-w&t=65s) [Agenda covering why agreements matter, why understanding them is hard, and the Docusign and NVIDIA approach.](https://assets.ai.engineer/sites/aiecode2025/sha256/e12258cdd2af17a30aab6f69436c0589f692cea61589a01721cc4ed7a15bf28c.webp)

What we’ll cover today:

Why agreements matter
• Why: Understanding them is hard
• How: Docusign & NVIDIA are working to solve for this

### Agreements Power Every Enterprise Business

Docusign’s 1.9 million customers and 1 million documents processed daily show the scale of making agreement data queryable. The slide says traditional tools destroy tables containing critical data.

[1:36](https://www.youtube.com/watch?v=_gvamfT8H-w&t=96s) [Slide lists customer, country, daily document, and agreement management cost figures alongside a partially cropped Docusign platform graphic.](https://assets.ai.engineer/sites/aiecode2025/sha256/017318829c586de6c5fa18ad98a1562f647c13de88db1da5d59c6038cd4b2f1e.webp)

Agreements power every enterprise business

But the most critical data is trapped in tables that traditional tools destroy.
At Docusign's scale this a structural blocker to agreement intelligence.

1.9 Million
Customers

180+
Countries

1 Million
Documents processed daily

$2 Trillion
Poor agreement management cost

Docusign IAM platform diagram:
Customer Experience
HR
CLM
Sales
Procurement
Legal
Create
Commit
Manage
Agreement Preparation
Agreement Desk
AI-Assisted Review
Identity Verification
eSignature
Workspaces
Obligation Management
Reporting & Analytics
TRUST
docusign IAM
Platform
Workflow Builder — No-Code Automation
Agreement Manager — Intelligent Repository
App Center & APIs — Integrations
docusign | iris

### Understanding, Managing, and Searching Contracts

Agreements contain obligations, pricing and other terms, but answering questions about expirations or purchased tokens can require searching across related agreements.

[3:00](https://www.youtube.com/watch?v=_gvamfT8H-w&t=180s) [Illustrated vendor agreement surrounded by categories such as obligations and pricing, beside example questions about contracts.](https://assets.ai.engineer/sites/aiecode2025/sha256/32c2ea34e4276e8178011e9a47a2c042f685dc32d423f09ff1bc5127891449cd.webp)

Understanding, managing, & searching contracts feels nearly impossible

Vendor Agreements
SLAs
Obligations
Important Terms
Pricing Tables
Contract Risk
Status

Which contracts we have signed contain a force majeure clause?
What contracts are expiring this month?
How many tokens have we bought from ABC vendor?
Who do I notify when I have a data breach?
What is my pricing table?

### The Pain of Table Extraction Today

Line-by-line extraction can break table structure when cells are merged or borders are missing. Teams then spend hours digging through agreements to find answers.

[3:59](https://www.youtube.com/watch?v=_gvamfT8H-w&t=239s) [Diagram lists table extraction challenges, including merged cells, missing borders, multi-page tables, and reading order, alongside potential consequences.](https://assets.ai.engineer/sites/aiecode2025/sha256/853980ce2c1700f4ff8f1d9cf0d35c7ca9d229d371a19cb49436032afa199e91.webp)

The pain of table extraction today

Where Traditional Extraction Fails
No visible borders
Merged cells and nested layouts
Mixed formatting in global regions
Multi-page tables
Reading order confusion
Semantic blindness

Consequences at Scale
MISSED SLAS
OBLIGATIONS
DELAYED RENEWALS
END DATE
UPDATED CLAUSES
NOTICE PERIOD
CONTRACT RISK

### Docusign and NVIDIA Table Extraction Architecture

After rasterization, layout understanding routes tables through Nemotron Parse and text blocks through MuPDF and reading-order preservation, then merges the results.

[4:44](https://www.youtube.com/watch?v=_gvamfT8H-w&t=284s) [Architecture diagram separates table and text processing after document layout understanding, then combines extracted text blocks and tables.](https://assets.ai.engineer/sites/aiecode2025/sha256/cdaebad1c5a8ee62243de8b12efee38f00988522bf26970250b72148de5f486a.webp)

Docusign + NVIDIA: A modern table extraction architecture

Scalability
Layout understanding
Accuracy

Multimodal Enterprise Documents

Rasterization
Text
Table

Layout Understanding Model

Tables Processing
Table Blocks
Crop Tables from Blocks
Nemotron Parse
LaTeX output
LaTeX to JSON Convertor

Text Blocks Processing
Text Blocks
MuPDF Text Fill
Preserve Reading Order

Extracted Text Blocks & Merged Tables

### NVIDIA Nemotron Retriever for document intelligence

Nemotron Retriever and cuVS help find relevant documents; Nemotron Parse and Nano VL then extract and reason about information within them. The slide highlights Parse’s table extraction and Nano VL’s multimodal throughput.

[5:13](https://www.youtube.com/watch?v=_gvamfT8H-w&t=313s) [Overview of document retrieval and understanding using Nemotron Retriever, cuVS, Nemotron Parse, and Nemotron Nano VL.](https://assets.ai.engineer/sites/aiecode2025/sha256/f70750290d245d50a65e57c6a6b34f7fe81c3d40bf6e36eb62e7c69cdc811601.webp)

NVIDIA Nemotron Retriever empowers AI-driven doc intelligence
Efficient, high-accuracy extraction, retrieval and reasoning

Document Retrieval
Nemotron Retriever
35% better accuracy
Information Retrieval

NVIDIA cuVS
7x higher throughput
Vector Indexing

Document Understanding
Nemotron Parse
83% more accurate table extraction
Document Extraction

Nemotron Nano VL
Up to 6X higher throughput
Multi-modal
Reasoning

Find the right documents and Find the right information within the document

### Nemotron open resources for RAG

NVIDIA pairs open RAG resources, from models and data to blueprints and skills, with retrieval and embedding leaderboards. The speaker says the team shares model weights and techniques.

[6:37](https://www.youtube.com/watch?v=_gvamfT8H-w&t=397s) [Models, data, libraries, training scripts, blueprints, and skills above three retrieval and embedding leaderboards.](https://assets.ai.engineer/sites/aiecode2025/sha256/efe9e2391f1bf0e7484838227b3816762993c709da9b957e0909481e3ec1e497.webp)

Nemotron Open Models, Data, Libraries for RAG
Build information retrieval systems, fine-tuned with open data, and customized for your domain

Models
Data
Libraries
Training Scripts
Blueprints
Skills

Leaderboards

ViDoRe V1, V2, V3
Visual Document Retrieval

MTEB
Massive Text Embedding

MMTEB
Massive Multilingual Text Embedding

### Nemotron Parse VLM architecture

Nemotron Parse passes a document through a vision encoder, compression layer, and transformer decoder to extract structured text, bounding boxes, and classes. The speaker describes it as a single-shot extractor.

[7:06](https://www.youtube.com/watch?v=_gvamfT8H-w&t=426s) [Nemotron Parse features and a diagram connecting a document image, vision encoder, compression layer, decoder, and structured output.](https://assets.ai.engineer/sites/aiecode2025/sha256/c547e0f2ce5c8449c1af5a218c99ae376bc5ebc2b667034ab3f05deef5304f3d.webp)

Nemotron Parse: Cutting-edge VLM tech

VLM: <1B parameters
Combines layout, OCR, and semantics
2x more accurate table extraction
Served via vLLM

Document example:
5.6. Compute Capability
[illegible]

Vision Encoder (CRADIO)
Compression Layer

Structured Text +
Bounding Box +
Class
Extraction Prompt

Transformer Decoder

Structured Output
[illegible] CUDA C++ Programming [illegible]
[illegible] These thread scope... [illegible]
[illegible] 5.6. Compute Capab... [illegible]
[illegible] The compute capabi... [illegible]
[illegible] The compute capabi... [illegible]
[illegible] Devices with the s... [illegible]
[illegible] The minor revisio... [illegible]
[illegible] Turing is the arch... [illegible]
[illegible] CUDA-Enabled GPUs ... [illegible]
[illegible] Note: The compute ... [illegible]
[illegible] 5.6. Compute Capab... [illegible]
[illegible]

### Agreement Manager document list

Agreement Manager lists completed documents alongside their parties, status and agreement type, with keyword search and filters. The presenter introduces the repository and an upcoming upload.

[8:37](https://www.youtube.com/watch?v=_gvamfT8H-w&t=517s) [Docusign Agreement Manager shows a searchable list of completed agreements with parties, status, and agreement types.](https://assets.ai.engineer/sites/aiecode2025/sha256/d5731a123302ee22f52f9238f3d98f847d53b37cef4236a27a33bf7324414c27.webp)

Completed Documents
AI-Assisted
Track obligations
21 HR agreements found
Show Insights
Try keywords or phrases
Filters

File Name | Parties | Status | Agreement Type
nda_agreement.pdf | Innovatech Solutions, Inc.; John Doe Consulting, LLC | Active; Expires 4/3/2028 | Non-Disclosure Agreement
ast_sci_data_tables_sample.pdf | WGBH Educational Foun... | Active | Handbook
Randon MSA.pdf | ElectroCircuit Corp.; Silicon Systems, Inc. | Active | Master Service Agreement
Space-Tourism_L4_Equipment_PO.pdf | Orbit-Leisure Corp.; Passenger-Safety Org | Active | Purchase Order
Space-Tourism_L3_Project_SOW.pdf | Orbit-Leisure Corp.; Passenger-Safety Org | Active | Statement Of Work
Space-Tourism_L3_Change_Order.pdf | Orbit-Leisure Corp.; Passenger-Safety Org | Active | Change Order
Space-Tourism_L2_Standard_NDA.pdf | Orbit-Leisure Corp.; Passenger-Safety Org | Active | Non-Disclosure Agreement
Space-Tourism_L2_Proposal_Final.pdf | Orbit-Leisure Corp.; Passenger-Safety Org | Active | Proposal
Space-Tourism_L2_Services_Agreement.pdf | Orbit-Leisure Corp.; Passenger-Safety Org | Active | Services Agreement
Space-Tourism_L1_Master_MSA.pdf | Launch-Facility, Inc.; Orbit-Leisure Corp. | Active | Master Service Agreement
Space-Tourism_L2_Operations_SLA.pdf | Orbit-Leisure Corp.; Passenger-Safety Org | Active | Service Level Agreement

Uploaded: View Job (under each file)
25 / Page
1 - 25 of 603

Interface text:
docusign
Home
Agreements
Templates
Insights
Admin
SP
Start
All Agreements
Drafts
In Progress
Completed
Deleted
Folders
All Parties
New
Employees
PowerForms
Bulk Send
New navigation

[8:51](https://www.youtube.com/watch?v=_gvamfT8H-w&t=531s) [Completed agreements list with search and filters; playback controls cover some rows.](https://assets.ai.engineer/sites/aiecode2025/sha256/5b795364adc51855904b69a097d8f7fe6fbcf5f6bd377943e01af5ee0f8c0edc.webp)

Completed Documents
AI-Assisted
Track obligations
21 HR agreements found
Show Insights
Try keywords or phrases
Filters

File Name | Parties | Status | Agreement Type
nda_agreement.pdf | Innovatech Solutions, Inc.; John Doe Consulting, LLC | Active; Expires 4/3/2028 | Non-Disclosure Agreement
ast_sci_data_tables_sample.pdf | WGBH Educational Foun... | Active | Handbook
Randon MSA.pdf | ElectroCircuit Corp.; Silicon Systems, Inc. | Active | Master Service Agreement
Space-Tourism_L4_Equipment_PO.pdf | Orbit-Leisure Corp.; Passenger-Safety Org | Active | Purchase Order
Space-Tourism_L3_Project_SOW.pdf | Orbit-Leisure Corp.; Passenger-Safety Org | Active | Statement Of Work
Space-Tourism_L3_Change_Order.pdf | Orbit-Leisure Corp.; Passenger-Safety Org | Active | Change Order
Space-Tourism_L2_Standard_NDA.pdf | Orbit-Leisure Corp.; Passenger-Safety Org | Active | Non-Disclosure Agreement
Space-Tourism_L2_Proposal_Final.pdf | Orbit-Leisure Corp.; [illegible] | [illegible] | Proposal
Space-Tourism_L2_Services_Agreem[illegible] | [illegible] | [illegible] | Services Agreement
Space-Tourism_L1_Master_MSA.pdf | [illegible]; Orbit-Leisure Corp. | Active | Master Service Agreement
Space-Tourism_L2_Operations_SLA.pdf | Orbit-Leisure Corp.; Passenger-Safety Org | Active | Service Level Agreement

Uploaded: View Job (under each file)
25 / Page
1 - 25 of 603

Interface text:
AIEngineer conference demo.mov
00:01
01:38
Start
All Agreements
Drafts
In Progress
Completed
Deleted
Folders
All Parties
New
Employees
PowerForms
Bulk Send
New navigation

### Upload an agreement

The demo selects an order form for upload, assigns the job a name, and shows Allow access as the next step.

[9:06](https://www.youtube.com/watch?v=_gvamfT8H-w&t=546s) [Upload documents dialog with one file selected and a Next button.](https://assets.ai.engineer/sites/aiecode2025/sha256/8c627172cc485fc502a73d76a7e33b0121afd7c121a33e674832848acbed9718.webp)

Upload documents

1 Select Files

Job name *
Give this upload job a unique name so you can easily find it.
Upload-1-2026-5-7_133843

1 file selected
You can select 999 more documents.

Add more files
Add

Next

2 Allow access

### Review upload jobs

The uploaded order form appears as a completed job. Its Actions menu offers Browse Files to open the agreement.

[9:14](https://www.youtube.com/watch?v=_gvamfT8H-w&t=554s) [Uploaded agreements job list with an Actions menu open.](https://assets.ai.engineer/sites/aiecode2025/sha256/fda19635f95c246c322047a43614eba9b96c5200a0d667e35b15fd51a08ae5fe.webp)

All jobs
View the status of any job, including uploads, AI processing, and synced agreements.

Uploaded

Uploaded Agreements
These agreements have been added to Navigator from a computer or third-party app. Track the progress for each job in the Status column.
Upload Files

Filters: Created date; Status; Source

Created Date | Job Name | Source | Files | Status
5/7/2026 11:38:58 AM | Upload-1-2026-5-7_133843 | UR-UI | 1 | Upload Complete; Documents processed with AI; 1 file available in Navigator
5/7/2026 11:32:48 AM | Upload-1-2026-5-7_133240 | UR-UI | 1 | Upload Complete; Documents processed with AI; 1 file available in Navigator
4/30/2026 8:07:45 AM | Upload-1-2026-4-30_10739 | UR-UI | 1 | Upload Complete; Documents processed with AI; 1 file available in Navigator
4/30/2026 8:02:55 AM | Upload-1-2026-4-30_10250 | UR-UI | 1 | Upload Complete; Documents processed with AI; 1 file available in Navigator
4/27/2026 4:28:56 PM | Upload-5-2026-4-27_182849 | UR-UI | 5 | Upload Complete; Documents processed with AI; 5 files available in Navigator
4/23/2026 4:08:11 PM | Upload-1-2026-4-23_1883 | UR-UI | 1 | Upload Complete; Documents processed with AI; 1 file available in Navigator
4/6/2026 5:31:07 PM | Sample Table | UR-UI | 1 | Upload Complete; Documents processed with AI; 1 file available in Navigator
4/6/2026 5:22:31 PM | Upload-1-2026-4-6_192226 | UR-UI | 1 | Upload Complete; Documents processed with AI; 1 file available in Navigator

Actions
Browse Files
View Details
Export Agreement Data
Delete

Interface text:
All Jobs

### Extract agreement tables

The demo highlights the order’s pricing table alongside three extracted tables. The presenter says NVIDIA’s Nemotron model parses the pricing data, which can be downloaded.

[9:26](https://www.youtube.com/watch?v=_gvamfT8H-w&t=566s) [Agreement viewer showing extracted tables alongside a highlighted order-details table.](https://assets.ai.engineer/sites/aiecode2025/sha256/af6a89fa27bd59ebc80981ab2bc617991540bd637cc45fbef31c7fee4590b7ae.webp)

SELLER
STELLAR LOGICAL
1855 Grant St.
San Mateo, CA 94402
United States

BUYER
FONTARA
500 Innovation Drive
Austin, TX 78701
United States

1. ORDER DETAILS

This Order Form ("Order") is entered into pursuant to the Master Services Agreement between Stellar Logical ("Seller") and Fontara ("Buyer"). The following products and services are ordered under the terms of that agreement:

Item Description | Quantity | Price | Total Price
AI Prototyping Licenses - Enterprise Edition | 10 | $500.00 | $5,000.00
Annual subscription, per-seat license
Implementation & Onboarding Services | 1 | $2,500.00 | $2,500.00
One-time setup and configuration
Premium Support Package | 1 | $1,200.00 | $1,200.00
24/7 priority support, 12-month term
Subtotal | $8,700.00
Tax (0%) | $0.00

Interface text:
Stellar_Logical_Fontara_Order_Form_100123.pdf
Download
1 / 3
100%
Details
AI-Assisted
General
Stellar_Logical_Fontara_Order_Form_100123.pdf
LINE OF BUSINESS
Unspecified
LANGUAGES
English
Show 6 empty fields
Termination
Renewal
Payment
Legal and Compliance
Tables
3 tables extracted
Download all
CONTRACT PRICING TABLE - LICENSES AND SERVICES
Found
PURCHASE ORDER HEADER
Found
LICENSE TIER DISCOUNTS
Found
Clauses

### Review discount tiers

Agreement Manager lists three extracted tables beside the order form. The renewal schedule shows the effective price falling from $500.00 for 1–9 licenses to $400.00 for 50+.

[9:45](https://www.youtube.com/watch?v=_gvamfT8H-w&t=585s) [Agreement viewer showing payment schedule and volume-discount tables.](https://assets.ai.engineer/sites/aiecode2025/sha256/c00877b67c76032bbf632e41eed20833532d1f270443311e9b20808bb6f36205.webp)

total amount paid under this Order Form.

Payment Schedule
Milestone | Due Date | Amount
Initial Payment (50%) | June 15, 2026 | $4,350.00
Final Payment (50%) | August 15, 2026 | $4,350.00

Volume Discount Tiers
The following discount schedule applies to AI Prototyping License renewals:
Tier | License Count | Discount | Effective Price
Standard | 1-9 | 0% | $500.00
Bronze | 10-24 | 10% | $450.00
Silver | 25-49 | 15% | $425.00
Gold | 50+ | 20% | $400.00

3. ACCEPTANCE

Interface text:
Stellar_Logical_Fontara_Order_Form_100123.pdf
Download
2 / 3
100%
Details
AI-Assisted
General
Stellar_Logical_Fontara_Order_Form_100123.pdf
LINE OF BUSINESS
Unspecified
LANGUAGES
English
Show 6 empty fields
Termination
Renewal
Payment
Legal and Compliance
Tables
3 tables extracted
Download all
CONTRACT PRICING TABLE - LICENSES AND SERVICES
Found
PURCHASE ORDER HEADER
Found
LICENSE TIER DISCOUNTS
Found
Clauses

### Nemotron + Docusign: Impact + Learnings

Docusign reports that Nemotron extracted tables 20x faster than other open-source models it tested. It favors smaller, purpose-built models for niche enterprise use cases, citing lower latency and cost.

[11:22](https://www.youtube.com/watch?v=_gvamfT8H-w&t=682s) [Slide reports model efficiency, 20 times faster table extraction, and the suitability of smaller models for niche enterprise use cases.](https://assets.ai.engineer/sites/aiecode2025/sha256/7009198226fe7958cced550870c8fbc8b10e024246bcd434c6da3ffc64aed4fd.webp)

Nemotron + Docusign: Impact + Learnings

Model efficiency
Fewer params (1B vs. 85B)

Faster extractions
20x faster per table

Smaller vs. Larger
Smaller models best for niche enterprise use cases

### What's next

Docusign and NVIDIA plan to move from extracting information to finding pages, deepening NeMo Retriever integration and exploring the NVIDIA Agent Toolkit, RAG applications, and production-scale agents.

[12:41](https://www.youtube.com/watch?v=_gvamfT8H-w&t=761s) [Slide lists deeper NeMo Retriever integration, a search platform roadmap, NVIDIA Agent Toolkit exploration, and RAG-based applications.](https://assets.ai.engineer/sites/aiecode2025/sha256/0f40a59a327a131188a714dd080e60c48b2f05cd1319057adf4844efd75cd26f.webp)

What’s next

NeMo Retriever deeper integration
Platform roadmap for search
NVIDIA Agent Toolkit exploration
RAG-based applications

### Docusign and NVIDIA

[13:13](https://www.youtube.com/watch?v=_gvamfT8H-w&t=793s) [Docusign and NVIDIA logos on a purple background.](https://assets.ai.engineer/sites/aiecode2025/sha256/cc62f936ce88dab218bb9bb2d0f5d24e2990e85c77f2bb487d3a00a8eaa897e4.webp)

docusign™
NVIDIA

### AI Engineer World’s Fair

[18:17](https://www.youtube.com/watch?v=_gvamfT8H-w&t=1097s) [AI Engineer World’s Fair title card on a black background, with the event website below.](https://assets.ai.engineer/sites/aiecode2025/sha256/f65496ff63a1f60c9b4260f0adaf3155bf0937c42c81e20d44bd3bf53fe50d71.webp)

AI Engineer
World’s Fair

HTTPS://AI.ENGINEER

## Full transcript

[0:13](https://www.youtube.com/watch?v=_gvamfT8H-w&t=13s) 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.

[0:21](https://www.youtube.com/watch?v=_gvamfT8H-w&t=21s) Hello, everyone. I'm a product manager at NVIDIA.

[0:23](https://www.youtube.com/watch?v=_gvamfT8H-w&t=23s) 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.

[0:54](https://www.youtube.com/watch?v=_gvamfT8H-w&t=54s) 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

[1:23](https://www.youtube.com/watch?v=_gvamfT8H-w&t=83s) 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

[1:53](https://www.youtube.com/watch?v=_gvamfT8H-w&t=113s) 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

[2:23](https://www.youtube.com/watch?v=_gvamfT8H-w&t=143s) 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

[2:52](https://www.youtube.com/watch?v=_gvamfT8H-w&t=172s) 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

[3:22](https://www.youtube.com/watch?v=_gvamfT8H-w&t=202s) 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.

[3:52](https://www.youtube.com/watch?v=_gvamfT8H-w&t=232s) 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

[4:22](https://www.youtube.com/watch?v=_gvamfT8H-w&t=262s) 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,

[4:52](https://www.youtube.com/watch?v=_gvamfT8H-w&t=292s) 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.

[5:17](https://www.youtube.com/watch?v=_gvamfT8H-w&t=317s) 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,

[5:47](https://www.youtube.com/watch?v=_gvamfT8H-w&t=347s) 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

[6:17](https://www.youtube.com/watch?v=_gvamfT8H-w&t=377s) 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

[6:47](https://www.youtube.com/watch?v=_gvamfT8H-w&t=407s) 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

[7:16](https://www.youtube.com/watch?v=_gvamfT8H-w&t=436s) 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

[7:46](https://www.youtube.com/watch?v=_gvamfT8H-w&t=466s) 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.

[7:59](https://www.youtube.com/watch?v=_gvamfT8H-w&t=479s) 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.

[8:30](https://www.youtube.com/watch?v=_gvamfT8H-w&t=510s) So that'll-- I think we believe we have a demo as well.

[8:32](https://www.youtube.com/watch?v=_gvamfT8H-w&t=512s) Yeah.

[8:33](https://www.youtube.com/watch?v=_gvamfT8H-w&t=513s) So how does I-

[8:34](https://www.youtube.com/watch?v=_gvamfT8H-w&t=514s) You just press one... Yeah.

[8:35](https://www.youtube.com/watch?v=_gvamfT8H-w&t=515s) Okay, there we go.

[8:36](https://www.youtube.com/watch?v=_gvamfT8H-w&t=516s) 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

[9:06](https://www.youtube.com/watch?v=_gvamfT8H-w&t=546s) 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

[9:36](https://www.youtube.com/watch?v=_gvamfT8H-w&t=576s) 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--

[10:06](https://www.youtube.com/watch?v=_gvamfT8H-w&t=606s) 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

[10:36](https://www.youtube.com/watch?v=_gvamfT8H-w&t=636s) 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

[11:06](https://www.youtube.com/watch?v=_gvamfT8H-w&t=666s) 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.

[11:36](https://www.youtube.com/watch?v=_gvamfT8H-w&t=696s) 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

[12:06](https://www.youtube.com/watch?v=_gvamfT8H-w&t=726s) 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

[12:35](https://www.youtube.com/watch?v=_gvamfT8H-w&t=755s) 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.

[12:45](https://www.youtube.com/watch?v=_gvamfT8H-w&t=765s) 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.

[13:12](https://www.youtube.com/watch?v=_gvamfT8H-w&t=792s) Perfect. I think that's what we had. We have time for a couple questions, anyone in the room.

[13:22](https://www.youtube.com/watch?v=_gvamfT8H-w&t=802s) Okay. Have someone there.

[13:24](https://www.youtube.com/watch?v=_gvamfT8H-w&t=804s) 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.

[13:45](https://www.youtube.com/watch?v=_gvamfT8H-w&t=825s) 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

[14:14](https://www.youtube.com/watch?v=_gvamfT8H-w&t=854s) 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

[14:44](https://www.youtube.com/watch?v=_gvamfT8H-w&t=884s) purpose? But I would let, let you add.

[14:47](https://www.youtube.com/watch?v=_gvamfT8H-w&t=887s) 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

[15:17](https://www.youtube.com/watch?v=_gvamfT8H-w&t=917s) spend that compute in that cycle will be changing between the different use cases.

[15:22](https://www.youtube.com/watch?v=_gvamfT8H-w&t=922s) Okay. One more there. Yeah.

[15:24](https://www.youtube.com/watch?v=_gvamfT8H-w&t=924s) 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-

[15:35](https://www.youtube.com/watch?v=_gvamfT8H-w&t=935s) 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

[16:05](https://www.youtube.com/watch?v=_gvamfT8H-w&t=965s) modules and stuff together. Yeah, one more. At last.

[16:09](https://www.youtube.com/watch?v=_gvamfT8H-w&t=969s) Uh, at scale, what quantization techniques are you using with the Nemotron model?

[16:14](https://www.youtube.com/watch?v=_gvamfT8H-w&t=974s) 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.

[16:22](https://www.youtube.com/watch?v=_gvamfT8H-w&t=982s) What about the Docusign? Are you, like, using the sixteen or are you quantizing down?

[16:26](https://www.youtube.com/watch?v=_gvamfT8H-w&t=986s) 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.

[16:34](https://www.youtube.com/watch?v=_gvamfT8H-w&t=994s) 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.

[17:05](https://www.youtube.com/watch?v=_gvamfT8H-w&t=1025s) Are you guys using, like, the, the category graphs on NVIDIA GPUs, or are you, like, primarily running through force?

[17:14](https://www.youtube.com/watch?v=_gvamfT8H-w&t=1034s) I believe they just deploy via vLLM directly.

[17:16](https://www.youtube.com/watch?v=_gvamfT8H-w&t=1036s) 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?

[17:29](https://www.youtube.com/watch?v=_gvamfT8H-w&t=1049s) Oh, th-this is just an extraction. This is not a retrieval.

[17:32](https://www.youtube.com/watch?v=_gvamfT8H-w&t=1052s) Yeah.

[17:33](https://www.youtube.com/watch?v=_gvamfT8H-w&t=1053s) Yeah.

[17:33](https://www.youtube.com/watch?v=_gvamfT8H-w&t=1053s) Oh.

[17:33](https://www.youtube.com/watch?v=_gvamfT8H-w&t=1053s) Yeah.

[17:34](https://www.youtube.com/watch?v=_gvamfT8H-w&t=1054s) Even when you're asking the question, was it a retrieval?

[17:35](https://www.youtube.com/watch?v=_gvamfT8H-w&t=1055s) It's coming from that agreement data that we've kind of extracted and stored.

[17:42](https://www.youtube.com/watch?v=_gvamfT8H-w&t=1062s) Yeah, maybe I can chat with you offline and how, like, we, our architecture kind of works fully as well.

[17:49](https://www.youtube.com/watch?v=_gvamfT8H-w&t=1069s) 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.

[18:01](https://www.youtube.com/watch?v=_gvamfT8H-w&t=1081s) Thank you.
