Authored by Shah. Explains Docusign’s collaborative integration of NVIDIA Nemotron Parse into its layout and OCR pipeline, serving through vLLM within Docusign’s environment and validation on real enterprise contracts. Beta availability, future general availability, agentic integrations and a public API are described as publication-time status or plans; authorship does not establish sole implementation responsibility.
Shah explicitly describes being part of this employer product launch, covering agreement questions, risks, renewals, approvals and customer-built agents through Agent Studio. The retained shortened announcement link does not establish her authorship of the announcement or sole product leadership.
Hiral Shah is a Senior Director of Product at Docusign, where she leads product strategy and development for agreement intelligence. Her work focuses on making the information buried in contracts usable: helping teams find obligations, compare prices, understand renewal terms, and act before important deadlines pass.
From product development to agreement intelligence
Shah holds a master’s degree in computer science from Carnegie Mellon University and an MBA from Stanford. Her career spans AI, mobile, and enterprise technology. Her Amplitude biography describes earlier work on first-generation Apple products and startup investing as a venture capitalist, experiences she connects to representing customers while defining product strategy.
Before Docusign, she was Director of Product Management at Amplitude, driving strategy for managing data at scale across its platform and partner ecosystem. Her published articles there covered proactive data governance, the Amplitude–Snowflake integration, and agile product management. That focus on organizing business data carries into her agreement-intelligence work: a signed document is useful only if the people responsible for its terms can retrieve and understand them.
At Docusign, the questions are concrete. An operations team responding to an outage needs to know which service-level notification requirements apply and whom to notify. Legal may need to find a contractor’s agreed hourly rate. Procurement may need to piece together pricing across several exhibits before renewing a vendor agreement. Answering those questions requires preserving the relationships inside a document, including the structure of its tables.
Turning contract tables into usable data
In her account of Docusign’s integration with NVIDIA Nemotron Parse, Shah explains why contract tables are a difficult production problem. Merged cells, nested layouts, inconsistent formatting, and tables spanning multiple pages can lead to incorrect extractions and manual correction. A rate or obligation loses its meaning if the system separates it from the heading or condition that explains when it applies.
Her account develops three priorities:
Reconstruct the table’s structure.Docusign integrated Nemotron Parse, a vision-language model built for document understanding, into its document-processing pipeline. The integration makes SLA obligations, contractor rate cards, and procurement pricing schedules available as structured, searchable data. Table extraction in Agreement Manager was accepting beta customers, with general availability still ahead, when she described it.
Keep confidential terms within Docusign’s environment. The company serves Nemotron Parse through vLLM and integrates it with its layout-detection and optical-character-recognition pipeline. This keeps sensitive agreement data inside Docusign while allowing its teams to run and optimize the model for their document-understanding needs.
Test on the contracts customers actually have.Docusign validated the integration against real enterprise contracts, including formatting variations, irregular table structures, and mixed-language content. These documents expose difficulties that synthetic benchmarks may miss; handling that variability is part of making extraction useful in everyday work.
Shah credits collaboration across product, engineering, and applied AI teams for bringing the capability to life. She also described presenting the joint work at AI Engineer World’s Fair with NVIDIA’s Sean Sodha, explaining how extracted agreement data becomes structured, searchable, and queryable information. Her published next steps included improving extraction accuracy on more varied tables, exploring agentic workflows through the NVIDIA Agent Toolkit, and preparing a public API for downstream integrations.
From renewal dates to agent behavior
Renewal management gives another example of how Shah approaches product development. In a public post about her work at Docusign, she describes being surprised that companies signing thousands of agreements could still struggle to answer a basic question: when does an agreement renew or expire?
She connects that customer insight to Renewal Management in Docusign IAM: using AI to identify renewal dates, understand how terms evolve, and give teams enough notice to act before termination deadlines. In the same post, she reported that her Agreement Renewal Framework patent had been approved and credited Docusign and her team. She also described the launch of a Renewal Agent, with a longer-term ambition for agents to manage renewals, help secure better terms, and identify opportunities beyond deadline tracking.
Her public reflections on AI-native product development extend this move from retrieving information to taking action. Following a Docusign session with Calibre Labs’ Justin Bauer and Sandhya Hegde, she highlighted ideas that resonated with her:
Build capabilities before the interface. Starting with MCP tools, reasoning, and orchestration makes the GUI one layer of the product experience rather than its starting point.
Design agents around reasoning and adaptability. Treating an agent builder as drag-and-drop workflow automation overlooks the behavior that makes an agent useful when circumstances change.
Make evaluation part of product development. Product managers can write evaluations and shape agent behavior, while designers focus on outcomes and engineers address orchestration and governance. She places trust and evaluation alongside model quality as enterprise AI matures.
These positions connect her agreement-intelligence work to a practical product question: once a system can recover a contract’s meaning, how should it help the customer decide and act? Table extraction supplies the structured information; renewal management shows why access to that information matters.
