Today Verneek, a New York-based generative AI startup, came out of stealth with the debut of its first product, Quin Shopping AI. The product is the first to utilize the company’s proprietary AI platform called One Quin.
The company, which was co-founded by the husband and wife team of Omid Bakhshandeh and Nasrin Mostafazadeh, spent the last two years developing the One Quin AI engine, which Mostafazadeh describes as a ‘consumer experience AI platform.’
“What we’ve done is that we’ve built a system which has many orchestrated modules of different transformer technology or non-transformer technology that has been trained to answer incoming questions,” said Mostafazadeh.
According to Mostafazadeh, the Shopping AI was trained with anonymously aggregated consumer query data gathered through the company’s initial partners (which she says she can’t reveal at this time) and synthetically-generated data sets based on these consumer queries.
Mostafazadeh said that the One Quin Shopping AI differs from other generative AI systems, such as ChatGPT, because it is vertically targeted around the specific use case of the consumer shopping experience.
“One Quin is AI plus curated knowledge in a box, whereas likes of ChatGPT is a general AI where knowledge is not curated.”
One benefit of this vertical focus is that, according to Mostafazadeh, their product will not suffer from the hallucination problems that plague general generative AI systems. General-purpose generative AIs like ChatGPT will sometimes produce answers that, while seemingly plausible, can be factually wrong or non-sensical. In contrast, One Quin is anchored by specific parameters within a confined topic set and is architected in a way in which it produces reliable answers.
“We’ve literally spent the last two years to mitigate that (hallucination),” said Mostafazadeh. “What is very unique about what we’ve created One Quin to sit on top of data. So it doesn’t generate off the wild. Instead, through very sophisticated inner machinery, it points to data that it sits on top of.”
Mostafazadeh said that because the One Quin engine is pointed to specific data, it can respond to specific questions tailored around parameters consumers use when searching for a product. For example, suppose a customer has a question about a food or nutrition product that fits a specific price range. In that case, One Quin can access this data and produce a tailored response specific to a retailer’s product inventory.
“What Quin can do, for example, is answer a question like ‘what is the healthiest snack I can buy for my kids that costs under $5?'” said Mostafazadeh.
I asked Mostafazadeh how her AI can determine whether a product fits criteria like healthiness, which can sometimes be arbitrary. She told me they had created something akin to a “health score” based on nutritional research. For other more arbitrary criteria, she told me the system is designed to anchor the answers with data points they believe act as a good proxy.
“For tastiness, Quin is basing it on the rating that the items have,” said Mostafazadeh.
Over time, however, Mostafazadeh says they could develop criteria to score a product for something like tastiness more accurately. However, one challenge with that, for now at least, is that the system is currently architected to answer questions without knowledge about the shopper.
“Right now, we have decided to make the barrier to entry basically zero. We don’t even ask the shoppers to log in. We don’t track them, and hence it’s a blank slate.”
That could change, said Mostafazadeh, who admits adding personal shopper contextualization would be very powerful.
“We would love to know that you are vegan without you telling me you’re vegan in your query. I would love to know that you hate cilantro because it tastes soapy, and by default, I will show you all recipes that don’t have cilantro in them.”
Mostafazadeh said that another advantage of Open Quin is that it can sit on top of any compute engine, whether it’s Microsoft Azure, AWS, Google Cloud, or in-store edge computing architecture. She said this makes it more affordable than other generative AI systems and gives retailers – who can be very specific about what cloud or computer system infrastructure they tie into – more flexibility.
“You probably know that retailers don’t like AWS (Amazon’s cloud). They don’t want anything of their world that touches anything of Amazon’s world.”
Mostafazadeh said that Quin Shopping AI could be deployed using various user interfaces. For example, she said retailers could deploy it in an app, on a website, via a chatbot, or on a consumer kiosk.
The company has raised a $4.2 million pre-seed funding round, and its website went live today.
Leave a Reply