Nvidia has announced the launch of a new blueprint for the development of generative AI-powered (genAI) retail shopping assistants.

The new blueprint is designed to help developers create AI-powered digital systems that work alongside and for human workers by offering the expertise of a retailer’s sales associates, stylists, or designers to their customers.

The US tech giant describes its ‘blueprints’ as reference workflows for both agentic and generative AI use cases.

At a press briefing, Nvidia’s director of product marketing for retail Cynthia Countouris described that two cross-industry AI blueprints announced by the company’s chief executive Jensen Wong earlier this week as the "starting blocks to develop custom agentic AI solutions", adding that the launch of the shopping assistant blueprint will “take this a step further”.

While genAI provides responses based on a single interaction using natural language processing, the next level of AI is agentic AI, which is able to solve more complex and multi-step problems through sophisticated reasoning and iterative planning.

Nvidia's retail blueprint provides an end-to-end multimodal capability which enables the assistant to understand text and images, while a multi-query capability means that the technology can search for several products at a time.

For example, the shopping assistant can be asked a general question about whether a retailer has tops or shoes, whilst also asking for recommendations for a purse to go with a particular skirt.

“Notice it also handles context of an ongoing conversation as well,” said Countouris. “In effect, you can ask for additional details on a recommended product such as, what is the heel height of those shoes? Or, how much can that purse hold?

“Notice I’m being very conversational, just as if I were working with a store associate or stylist. And you're delivering this through digital channels to your customers.”

In the home furnishing market, as well as facilitating furniture recommendations; product details such as fabric care instructions; and a search option to find products that are similar to an image of an item, the shopping assistant also allows customer to select furniture from a catalogue and see it represented in their own room.

This is made possible through Nvidia Omniverse, a real-time AI-based 3D graphics collaboration platform.

Talking about how the technology is able to provide a hyper-personalised experience, Countouris said: “It takes context both from how the customer is engaging right now with the store associate and the AI blueprint can very easily be extended with retrieval-augmented generation (RAG) to look at the customer's previous purchase history and engagement. It's a RAG-based model.”

On alignment with Nvidia's announcement on Friday, brand IT advisor SoftServe unveiled its genAI shopping assistant, which was developed using the new blueprint.

The company's assistant is able to facilitate virtual try-on, allowing customers to visualise how products look on them directly through an online chat before making a purchase.

Nvidia has witnessed several trials of the technology being used to provide interfaces for customers to virtually try on products in store, with this feature seemingly gaining more traction.

Countouris also revealed that the organisation is in discussions with several retailers in Europe about its new blueprint, with official announcements set for a later date.

Looking ahead, Countouris said that agentic AI will be transformative in the retail industry.

“...agentic AI just brings it another level further in helping to provide more tools and resources to workers in retail to be able to take on new tasks, do them more quickly and more effectively,” she added. “So this expansion is going to really revolutionise what's happening in retail.”

The director believes that the adoption of agentic AI is likely to grow over the next couple of years, predicting that the technology will not just be used in the back office for e-commerce but also for other areas, such as in the supply chain supporting warehouse and distribution workers, for example by identifying conveyor system breakdowns.

---