Inspiration
I have nothing to wear! I forgot my favorite red top was in the back of the closet! As victims to these thoughts every now and then, we decided to get rid of the hassle altogether! Introducing Neural Threads, your personalized AI stylist that will not only pick chic outfits for you but will also make sure you're making sustainable wardrobe choices!
What it does
Our two agents built using Gemini Vision API, React (Vite), REST APIs and SQLite act as your personal AI stylist and AI declutterer. You get to digitize your wardrobe, get live outfit suggestions based on your 'Mood of the day', and even get a detailed breakdown of clothes you haven't worn in ages! What's more? You can save your outfits and access them anytime! The star is our "Do I need this?" sustainability feature which refrains us from buying stuff we don't need - because we already have it!
How we built it
Frontend: React (Vite), TypeScript - website only, deployed on Digital Ocean. Backend: FastAPI in api/ - upload, classify, wardrobe, stylist, declutter. AI: Google Gemini (Vision + Text), Open-Meteo. DB: SQLite.
Challenges we ran into
It was challenging to get 'Monica' - our declutter agent to recognize which clothes weren't worn in ages so she could suggest the user to DONATE them.
State management between the two agents was one of our hurdles. We applied rules to enforce each of their roles to solve any conflicts.
The accuracy of parsing image data to JSON format through Gemini Vision was technically challenging.
Accomplishments that we're proud of
Our declutter agent and the "Do I need this?" feature - In the world of fast fashion and disappearing clothes in the closet - this will be a game changer! It will give you back your wardrobe space and you will give back so much more to the environment!
Open-Meteo feeds local weather data to our stylist, Miranda, who then curates outfits tailored to the conditions for you.
You never have to worry about going to Pinterest, finding outfits you like then going to your closet to rummage through tons of clothes to recreate them! It's done for you, according to your preferences, in the app!
What we learned
Working with Gemini Vision was really enlightening. Working with Gemini in an agent-style workflow taught us that prompting and orchestration matter as much as the model itself, especially when you want predictable outputs instead of random responses.
The biggest takeaway was that building AI features that feel polished is more about system design, user experience and iteration speed than just calling a model once.
What's next for Neural Threads
Once the user is accustomed to the web app, we would love to explore the expansion of Neural Threads into a mobile app for more accessibility and app-friendly features for our users.
Built With
- agents
- ai
- api
- digitalocean
- fastapi
- gemini
- geminivision
- open-meteo
- python
- react
- sqlite
Log in or sign up for Devpost to join the conversation.