-
-
Example: Dress
-
Example: Shoe
-
Example: Dress - Different model used
Inspiration:
The idea came from selling kids clothing online, and what a hassle the process of writing good listings can be for the seller. It became more real when new generative AI tools and API's could be put together in a simple way to create a solution, and make this process much faster, with better results, and with less burden on the user. We found some great market information from ThreadUp ReSale Report 2023.
What it does:
It writes really fun and effective listings for ReSale clothing and accessories. It generates and integrates useful SEO terms, hashtags, and manual inputs from dropdowns into a usable listing description for the person to review, edit, and list as they wish.
How we built it:
We built a web app with ReactJS, which accepted images to be uploaded, with different descriptors the user can select. Those images were sent to a Node.js Express server, which then leveraged the Google Vision API to receive JSON data about the image. That JSON data was then fed to OpenAI Chat Completion API (Chat GPT) to generate an advertisement based on the image data. The advertisement would then be returned to the React Client and shown for the user. A similar approach was also implemented using Python/Flask as the frontend but still using google vision API and Chat GPT with similar results
Challenges we ran into:
It was a challenge to determine the proper scope for the short hacking period of time, and when we completed what we thought we could, the time was extended, which was nice but brought up questions about our scope. Initially, lots of ideas on what this application could do and how it could be done had to be narrowed significantly. Also, we all just met for the first time making it difficult to know what each person can and should contribute. From the programming side, it was challenging to set up the google vision API to correctly output characteristics about the clothing. While we have an initial model going, we noticed there are some inaccuracies, and the AI model needs to be trained/improved to make the listings even better going forward.
Accomplishments that we're proud of:
We got the thing to work and integrated 2 AI platforms for a very usable output. We had 2 programmers working together, but on different coding language, and were able to achieve a proof of concept result. We were able to narrow down the scope into a marketable idea and platform showcasing both a social impact and economic opportunity. We stuck together as a team and each contributed uniquely from our perspective and background. We each tried new things in the process and had a good time doing it.
What we learned?
We learned that new generative AI tools are very effective as programming partners, and that vision AI's can return very useful analysis for certain applications. We learned you can connect them through a website with simple interface to achieve different outcomes. We learned its great to have a go with a diverse team, getting focused on solving a problem(s) to deliver a workable result in a constrained period of time. We learned the importance to figure out what makes our solution technically novel + valuable for the key customer. Finally, we learned that the Google and Open AI API's are simple to use, yet powerfully rich with their capabilities even this early in their life cycle's.
What's next for ReSaleWriter.AI
We have interest to continue development of this tool for commercial use and see a pathway to making a positive social impact. Lets see what happens at the Hackathon first, then pivot to the most appropriate, focused solution that's novel, technically possible, works well, and positioned right for the clearly identified customer.
Log in or sign up for Devpost to join the conversation.