MeteoroloChic

Fashion That Adapts to Your Forecast
Cover Page
Value Proposition
Development Process
Development Process Contd.

Inspiration

I was inspired to make this application due to the issues I experienced traveling to different places with my girlfriend. My girlfriend has serious issues deciding what clothing to buy for different seasons or when sharp and drastic weather changes occur and how to caption her photos as her field is English dominated and she is still learning English language.

What it does

This app is created as a dual-feature app, your personal stylist and shopping assistant rolled into one. Firstly, it curates a selection of outfits based on the weather in your specific location, be it your street or city. These outfit suggestions are generated using the Arctic Instruct model. If any of these catch your eye, you have the option to purchase them in-store. Secondly, the app allows you to create an outfit based on your own descriptions. It’s a simple and effective way to determine if an outfit suits your needs given the current weather conditions. But that’s not all! Our app also features a caption generator which can generate engaging captions for your Instagram photos, again powered by the Arctic Instruct model.

For advertisers, this app offers a unique opportunity. It allows them to showcase their closest matching clothing items to the generated image. This targeted advertising feature can be seamlessly integrated into the application, paving the way for revenue generation.

In essence, this app is not just a fashion tool, but also a platform for discovery and inspiration. Enjoy the convenience of having a personal stylist, shopping assistant, and social media aide, all in the palm of your hand!

How we built it

This application was built with streamlit, snowflake, python, huggingface, replicate programming language and libraries respectively. Weatherapi and Opencage api are the application programming interfaces which are chosen for the tasks of generating accurate latitude and longitude scores for passing into the weather api as I wanted to be able to resolve street address level data and weatherapi could only do city level information. Huggingface and replicate inference models were used for making inference calls and returning relevant information given inputs.

Challenges we ran into

I found out about this competition about 1 week to completion so I had to quickly brainstorm on a problem and solution. There are at most 2 or 3 repositories out there on image captioning which are functional and none on weather fashion recommendation which are functional. So there is little to no information to access on this and calibrating the captions and models without importing torch or Tensorflow.

As a recent graduate of social work from the university of Lagos I had so much to learn about the different API´s and how to connect them and have them return the right output. I had to use the output of one model as input for another model and 3 models are utilized in total.

Finally, the snowflake arctic instruct model has a weird tendency to generate responses with the word 'none' appearing throughout the text response. I had to clearly specify that it avoid the word 'None' in my prompt. The arctic model also had a tendency to generate responses which weren't relevant until I started calibrating it to the kind of inputs I wanted. If I had time I would have implemented an RAG to help it better understand what I wanted it to do.

Also, the output from the replicate inference call for the snowflake arctic instruct model is returned broken into different new lines and I have had to work very hard to resolve this issue.

Accomplishments that we're proud of

I am incredibly proud of the strides I have made in such a short time. Despite the steep learning curve and the limited resources available, I managed to create a functional application that not only solves a real-world problem but also has the potential for revenue generation. I successfully integrated multiple APIs and models, a feat we initially thought was beyond our reach. This project has truly been a testament to my resilience, adaptability, and determination.

What we learned

This journey has been a profound learning experience. I learned about various APIs, programming languages, and libraries. I discovered the intricacies of connecting APIs and ensuring they return the correct output. I also learned how to calibrate models without importing certain libraries. More importantly, I learned that with perseverance and a willingness to learn, I can overcome any challenge that comes my way.

What's next for Photo Caption Generator and Weather Garment Recommender

I am excited about the future of this application. I plan to add multi-language support to make our app more inclusive and user-friendly. I also aim to improve the UI for a better user experience. Furthermore, I intend to add a section where advertisers can display targeted ads based on the images generated by the model. This will not only enhance the functionality of the app but also open up new avenues for revenue generation. We look forward to continuing our journey of learning and innovation.