NVIDIA launches ‘easy button’ for creating gen AI workflows

Aiming to provide businesses with an “easy button” for starting development of enterprise generative AI applications, NVIDIA on Tuesday unveiled NVIDIA NIM Agent Blueprints, a catalog of pretrained, customizable AI workflows for a host of use cases, including customer service avatars, retrieval-augmented generation (RAG), and drug discovery virtual screening.

“NVIDIA NIM Agent Blueprints are runnable AI workflows pretrained for specific use cases that can be modified by any developer,” said Justin Boitano, vice president of enterprise AI software products at NVIDIA.

“This is an ever-growing catalog of reference applications built for common use cases that encode the best practices from NVIDIA’s experiences with early adopters,” he added.

The catalog is built on NVIDIA NIM, a slate of microservices composed of downloadable software containers for speeding the deployment of enterprise gen AI applications.

Boitano explained that for years enterprises have relied on teams of developers to create and maintain custom, purpose-built, in-house applications to run their core business processes.

“You can think of these applications as a database wrapped in a web UI connecting multiple teams through a business process,” he said. “We realized it’s time to really scale the impact of generative AI by enabling these tens of millions of enterprise app developers to create this new form of in-house, enterprise application in a way that’s easier for them to get going.”

Sample applications in the catalog are built with NVIDIA NeMo, NVIDIA NIM, and partner microservices, reference code, customization documentation, and a Helm chart for deployment. Enterprises can modify the sample applications using their own business data and run the resulting gen AI applications across accelerated data centers and clouds. The blueprints are free for developers to download and can be deployed in production with the NVIDIA AI Enterprise software platform.

Customer service, PDF extraction, drug discovery

NVIDIA’s initial NIM Agent Blueprints offering covers three use cases: a digital human workflow for customer service, a multimodal PDF data extraction workflow for enterprise RAG, and a generative virtual screening workflow for computer-aided drug discovery.

“We focused on first launching these because we think they are super critical for every industry,” Boitano said. “But we’re going to have a roadmap of more workflows coming out on a monthly basis.”

The digital human workflow for customer service is aimed at helping enterprises bring their enterprise applications to life with a 3D animated digital human interface powered by NVIDIA Tokkio, an interactive avatar virtual customer service assistant product SDK. Intended to help developers create customer service applications with “approachable, humanlike interactions,” the blueprint features NVIDIA software, including NVIDIA ACE, NVIDIA Omniverse RTX, NVIDIA Audio2Face, and Llama 3.1 NIM microservices. It’s designed to integrate with existing enterprise gen AI applications built using RAG.

“We started with digital human because we think it can inspire everybody to understand what a chat interface could look like in the future,” Boitano said. “They don’t all need to be text-based, question-and-answer. They can start to look and feel like people that you’d want to interact with more naturally.”

The multimodal PDF extraction workflow blueprint is intended to help enterprises quickly give their digital humans, AI agents, or customer service chatbots expertise by granting them access to the enterprise’s corpus of PDF data. Developers can use the blueprint to combine NVIDIA NeMo Retriever NIM microservices with community or custom models to build multimodal retrieval pipelines.

“There are trillions of PDFs generated every year across enterprises, and these PDFs include multiple data types, including text, images, charts, and tables,” Boitano said. “This is a goldmine of data that can be used as quickly as humans can read and understand it. By combining PDF data extraction and generative AI applications, this untapped data can be used to uncover business insights that can help employees work more efficiently.”

The generative virtual screening workflow for computer-aided drug discovery leverages NIM microservices including AlphaFold2, MoIMIM, and DiffDock, to help researchers and developers customize and deploy AI models for 3D protein structure prediction, small molecule generation, and molecular docking.

“Generative AI has been associated with understanding and generating human language, mostly,” Boitano said. “However, its potential extends far beyond human language, encompassing the complex languages of biology and chemistry. Proteins, which are the building blocks of life, have their own alphabet, comprising 20 amino acids, and generative AI models such as AlphaFold and MoIMIM are trained to interpret the language of proteins and chemicals, aiding in the drug discovery process.”

Additional blueprints to come

NVIDIA plans to release additional blueprints on a monthly cadence, including blueprints for gen AI applications for customer experience, content generation, software engineering, and product research and development. Boitano says blueprints for enterprise search and supply chain “what if” scenarios are in the works, along with blueprints geared to industry verticals such as manufacturing and retail.

NVIDIA is tapping into its partner ecosystem of system integrators, technology solutions providers, professional services providers, and server manufacturing partners to make NIM Agent Blueprints available. These partners include Accenture, Cisco, Dell Technologies, Deloitte, Hewlett Packard Enterprise, Lenovo, SoftServe, and Worldwide Technology (WWT).

“There’s going to be a virtuous cycle of OEMs working with GSIs [global systems integrators] with data science teams to drive business outcomes into production quickly, and, ultimately, I think our whole ecosystem is aligned around getting applications and generative AI into production faster,” Boitano said.