Savvy - AI Shopper

Shop for your best products through personalized insights and recommendations.

Comment

Inspiration

The modern online shopping landscape is broken. One of the foremost issues is the overwhelming abundance of choices. With tens of thousands of products available, each with slight variations, consumers face a sizable task in identifying the best options that meet their specific needs.

With this much information at our fingertips, the sheer volume of available data hinders decision-making rather than facilitating it. Consumers find themselves navigating through a maze of products, reviews, and features, which can be both time-consuming and confusing. The process is further complicated by the reliability and authenticity of online reviews and recommendations, which are crucial in guiding purchase decisions but can often be misleading.

The impact of these challenges is felt most acutely by non-savvy shoppers. These consumers tend to make suboptimal choices, either overpaying for features they don't need or settling for products that don't fully meet their requirements. As a result, they may end up spending more time and money than necessary.

Take, for example, a parent who is trying to buy a computer for their child. They often will buy whatever catches their eye first without regard to the features that make a computer powerful. Or, consider a more savvy but equally naive parent might ask online forums for help choosing. These two cases result in increased time spent and/or worse deals for the modern day shopper. We believe that the vast majority of shoppers are not savvy and thus inefficiently use their funds. We envision a world where non-savvy shoppers get the same or better outcomes as savvy shoppers without the additional time spent on shopping.

What it does

Our project is a product recommendation search engine named Savvy. Savvy aggregates the data of e-commerce retailers (currently only Walmart but others can be easily added as well), and produces a list sorted by how relevant they are to the shopper's query. For each option, a Savvy Score is given, along with insights about why it should be relevant to the shopper.

As a startup, we plan on generating revenue from affiliate links so that we can offer users a free experience.

How we built it

The flow looks like this:

  1. The user's query is fed into Google's Vertex AI, using PaLm 2 to parse their query into a Walmart-searchable query. For example, "My friend likes blue dresses. I want to get them one in size medium as a gift." -> "medium blue dresses"
  2. Query Walmart API's search functionality with the parsed query
  3. Clean the data into only the relevant information needed to produce insights, such as product description and ratings.
  4. Take that data and together with the shopper's initial query, give each Walmart product a ranking according to how well it fits the user's demands, presenting the data in a clear and concise way to reduce information overload.

Challenges we ran into

The first issue we ran into was finding the right online retailers as data for Savvy to use. We wanted to support Amazon first, but their API is opened to only to established Affiliate partners. We're taking steps to gain access to their API, which we expect approval for in the near future. We settled on the Walmart API but that too was also a challenge, as its endpoints required encrypted credentials, and the only option for encrypting our credentials was a piece of Java code on their website that we had to rewrite into Go.

At the beginning, we also struggled with getting the results we wanted from Google's Vertex AI as new AI engineers. However, through attending prompt engineering workshops, we gained insight on how to craft a detailed and effective prompt to yield responses that were up to scratch.

Accomplishments that we're proud of

  • Implementing Vercel's AI SDK to query Google Vertex AI through the Langchain provider
  • Figuring out how to integrate the with the Walmart API

What we learned

  • Developed our own AI Chatbot for our custom website
  • Craft effective prompts to control the behavior and output of AI models

What's next for Savvy

  • Become affiliates for other retailers such as Amazon to increase the amount of products that can be discovered through Savvy. Additionally, we will be able to monetize our product from commissions earned through affiliate links.
  • Improve the UI/UX, refine prompts
  • Do all of this in public to generate excitement for the product and to get and build on feedback

Built With

Share this project:

Updates