Tailor

Inspiration

Manually crawling sites for online shopping no longer made sense in the age of AI.

Our inspiration for Tailor stemmed from the frustration we all experience when trying to find the exact product we're envisioning online. We recognized the need for an interactive and precise tool to aid online shoppers, enabling them to dictate their preferences and receive accurate, personalized product suggestions in return.

What it does

Tailor is a browser extension that enables shoppers to flag and prompt the visual features of a product that they desire in their final purchase. Tailor then displays real products from across the internet that match those desired features, giving the user ultimate precision in finding exactly what they want.

How we built it

Tailor is a Chrome Extension built on NodeJS and a vanilla JS UI. Hosted on Replit with Express. Our core AI functionality is accessed via an unofficial Midjourney API. The resulting product description is sent to our API to query against a Weaviate vector database that we populated by scripting against the Etsy API. Some of our data was manually added with the help of Diffbot, an AI-powered web crawler/extractor. We developed a matching algorithm (in Python) whereby a user’s desired traits, undesired traits, and text-communicated preferences are the recipe for a match score. We return products with a match score above a certain threshold. Due to time constraints, we outsourced the semantic search to GPT4 as opposed to a custom-built NLP solution for the matching.

Challenges we ran into

The sheer volume of products on the internet presents an obvious challenge for which we had to find and gain access to bespoke solutions, e.g. web crawlers. The lack of a Midjourney API meant we had to find a workaround. Our team had to make up for the lack of a full-stack developer, so learning how to build a web-based product was a new challenge.

Accomplishments that we're proud of

We are particularly proud of our ability to transform the online shopping experience, making it more user-centric and less time-consuming. Our achievement lies in using AI to provide a highly personalized and relevant product recommendation system that spans the entire internet, connecting users to products they would have never discovered otherwise.

What we learned

We learned about Weaviate vector databases, Chrome extensions (hard), Replit hosting, webhook architecture, Approximate Nearest Neighbor searching, API integration.

What's next for Tailor

Refining the search results using more bleeding edge AI tech such as OpenAI CLIP, Facebook SAM, Stable Diffusion. We’ve brainstormed a novel search method where users prompt their preferences to iteratively generate their dream product using Stable Diffusion, and then use image similarity search like Google Lens to find products that match in the real world. We plan to deploy a mobile app. We intend to partner with major retailers for data sharing, and fashion influencers and bloggers to showcase Tailor's features and benefits to a wider audience.

Built With

• diffbot
• google-colab
• javascript
• json
• midjourney
• node.js
• python
• replit
• serpapi
• weaviate
• wix