Accelerated computing, AI infrastructure and graphics

NVIDIA

NVIDIA develops accelerated computing hardware and software for AI, scientific computing, graphics and autonomous machines. Its CUDA platform lets developers use GPU parallel processing for scientific simulations and AI model development, while NeMo supports custom generative AI, including speech recognition and synthesis. GeForce RTX serves gamers and creators; Jetson and Isaac help teams develop and deploy robots and edge AI applications across manufacturing, logistics, healthcare and retail.

Founded in 1993 by Jensen Huang, Chris Malachowsky and Curtis Priem, NVIDIA began with a focus on 3D graphics for gaming and multimedia. Huang remains CEO. The company’s ray-tracing research also contributed to RTX hardware, while its DLSS technology uses AI to reconstruct high-resolution images from a fraction of the rendered pixels.

NVIDIA’s products span cloud, data-center, desktop and edge deployment. DGX Spark supports local AI applications, including personal agents that can switch between local and cloud models. For industrial software developers and manufacturers, its simulation tools support physically accurate digital twins for building, training and testing systems before deployment. Omniverse extends this work with camera, lidar and radar simulation that developers can integrate into existing applications.

Explore the recordings

This NVIDIA archive currently contains one recording, co-presented with Docusign. It offers a practical path into agreement extraction, structured document data and enterprise workflows. The descriptions below reflect the recording; they do not establish current product capabilities or availability.

Start with the agreement workflow

Watch Your Agreements Are a Database You Can't Query for the business problem and demonstration: turning agreement pricing tables into queryable data while preserving tiers, SKUs and rate-card structure. Hiral Shah and Sean Sodha walk through order-form upload, metadata and pricing extraction, CSV export and API access, then discuss organization-wide agreement modeling and search. This is the most useful entry point for readers exploring how document extraction fits into an enterprise workflow.

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

Follow the extraction architecture and its limits

Return to the same recording for the presenters’ description of Nemotron parse: a small vision-language extractor returning text, layout, reading order and table structure in one pass, with NVIDIA NIM or vLLM serving options discussed. The Q&A is useful for understanding Docusign’s reported hybrid pipeline, which retains OCR for other fields and clause text, and the distinction between bulk preprocessing and interactive document Q&A. Shah reports roughly 20× greater table-extraction throughput than other open-source models they tested, but the supplied account lacks full comparison conditions. Treat that as a reported result, not a general benchmark. Quantization, multi-token generation, retrieval and agent work are discussed as future directions in the recording, not verified present-day capabilities.