Flux | Devpost

Chrome Extension
Login/Register Page
Homepage

Inspiration

FLUX arose from a common frustration: the overwhelming nature of online shopping and the time spent scrolling through different e-commerce platforms. Like Doc Brown's DeLorean time machine, we wanted to create something that could help users navigate the landscape of e-commerce with precision. The name "FLUX" pays homage to the iconic Flux Capacitor while representing our mission: Financial Learning & User eXperience.

What it does

FLUX is a chrome extension and web platform that acts as your shopping time machine. Our chrome extension tracks your e-commerce browsing patterns across multiple platforms, including Amazon, Target, EBay, and more. FLUX uses AI to analyze your shopping behavior and preferences and provides personalized product recommendations. Based on your shopping patterns, our AI is also able to provide a plan of action: a way to help you plan out your financial life in a smarter way. FLUX is user appealing: it offers a retro-futuristic dashboard that allows you to have a centralized place to view your financial habits.

How we built it

Our tech stack embraces modern solutions while maintaining the common theme: "Back to the Future":

Frontend: React with a "Back to the Future" inspired UI and a lightweight chrome extension
Backend: Node.js and Express
Database: MongoDB
AI/ML: llama3-8b with fast inference using Groq

Multi-Agent Framework for Most Useful Financial Advise

The system utilizes LangChain and Groq to orchestrate a sequence of specialized agents that output precise JSON formats, ensuring seamless internal communication. It begins by fetching user data (financial goals and purchase history) from MongoDB, then extracting brands and categorizing products to analyze spending patterns. Category and brand analysis agents generate detailed insights, while computation functions calculate key financial metrics such as total spending and purchase frequencies. Graph explanation agents convert these metrics into clear visuals, and a synthesis agent combines all insights into comprehensive, actionable advice. Inspired by recent advancements in AI inference optimization, this multi-agent workflow leverages specialized models for enhanced accuracy, modularity, and scalability, outperforming a single monolithic model by delivering more tailored and effective financial guidance. (See the agents workflow diagram in the Project Media).

Challenges we ran into

This project was our first time building a chrome extension. We had to ensure that our extension worked consistently across different e-commerce websites due to different platforms having different structures. Along with that, we came across difficulties in syncing up all of the data points and using them as inputs into our AI for processing. We were able to utilize LangChain's structured output to help with this and solve for this challenge. Additionally, creating a custom UI and connecting it to our model's output was a keen struggle too.

Accomplishments that we're proud of

We were proud to have created a full functional Chrome extension that tracked a user's activity on several e-commerce platforms. We were able to implement a robust AI recommendation system that provided the user with personalized suggestions as well as a realistic plan of action for their financial stability. We designed an engaging retro-futuristic UI that made shopping experience fun and accessible.

What we learned

We took a deep dive into browser extension architecture, gained practical experience with LLMs and recommendation systems, created themed interfaces that were both functional and engaging, and learned how to balance out our frontend and backend needs.

What's next for Flux

For the future of FLUX, we are thinking of adding more complex functionality in terms of more complicated financial goals, tailoring towards people's individual needs, and continuing to support more e-commerce platforms.