CleanBucket is an innovative app where users can record themselves recycling in a fun, engaging way and share their efforts with a community. People can scroll through videos, rate them, and get inspired by others to make recycling more exciting and impactful.

• Inspiration
• What It Does
• How It Works
• How to Use
• Challenges
• Accomplishments
• What We Learned
• What's Next

The idea for CleanBucket emerged from a desire to make recycling more fun and interactive. Recycling can feel like a chore for many, so we wanted to change that by making it more engaging. By combining social media elements with sustainability, we envisioned a platform where people could creatively showcase their recycling efforts and inspire others—similar to the popularity of short-form content on social media. The ultimate goal is to motivate more people to recycle by making it social, fun, and rewarding.

CleanBucket allows users to post videos of their recycling activities, from showing off impressive trickshots to just having fun with the process. The app includes a rating system to add a competitive edge where users can interact with others' videos, providing ratings and feedback. The app's recommendation algorithm ensures that users see the most engaging videos based on their preferences, utilizing multiple AI models to rank content.

1. Recording a Video:
   Users open the app and upload a video of themselves performing a recycling action, like a trickshot into the recycling bin or creatively throwing away trash.

2. AI Analysis:
   The video is analyzed using the LLAVA 1.5-7b-hf model hosted on CloudFlare's AI Workspace, which generates a detailed description of the content.

3. Agent Interaction:
   The description is processed by Fetch.ai agents running the TinyLLaMa-1.1b-chat-v1.0 model. These agents communicate with each other, generate comments, and provide feedback on the video.

4. Final Rating:
   A final AI model evaluates the conversation and descriptions to generate an overall rating for the video. This rating helps rank videos, and the best content is shown to other users.

5. Data Storage and Communication:
   All user data and interactions are stored using MongoDB Atlas, while video content is stored on a secure cloud server.

Once you clone the repo, you need to do deal with the Xcode a little bit. If not set already, go to Signing & Capabilities and see App Transportation Security Exception, then add video.cleanbucket.co. This is needed because iOS requires HTTPS. You might also have to go to General -> Targets -> CleanBucket -> Frameworks, Libraries, and Embedded Content and add Alamofire! Once you've done so, you have two options to use our little project: run it on a simulator or run it on your phone! The simplest is the simulator, which you can build and launch with the top left button; however, it is very slow so we suggest using your own device. You can connect your phone to your computer and change the simulator target to your phone, which can be seen at the center top of Xcode (DISCLAIMER: you may need to use dev mode for this). Then, Voilà!

Developing in Swift was one of our biggest challenges due to tight timelines and numerous technical hurdles. The amount of issues we encountered with Swift was comparable to all other platforms combined. However, overcoming these issues allowed us to get closer to our goal and learn a lot in the process.

• Successfully setting up an AI pipeline involving multiple layers of models and agents.
• Getting significant help from mentors, Fetch.ai, and CloudFlare, which gave us insights into the capabilities of AI agents working together.
• Developing a unique application that combines sustainability and social media elements.

Building CleanBucket from scratch was an immense learning experience. While it's too much to list everything, key lessons include:

• UI design: Next time, we would dedicate more time to crafting a more intuitive and user-friendly interface.
• Data flow: We learned the importance of thinking about how data will move across platforms and devices before diving into coding.

• User Feedback Integration:
  We would love to integrate user feedback directly into the AI pipeline, allowing users to rate the "coolness" of videos. This would help finetune the final rating system and improve the overall accuracy of predictions.

• Video Playback Optimization:
  We are looking into improving the video playback experience by reducing delays and improving the performance of the app, ensuring smooth interactions for users.