EcoShop

Inspiration

As young people who are conscious about protecting our environment, we are deeply interested in ensure that the purchases that we make are sustainable. However, by using simple google searches, we felt lost in a sea of vague claims and greenwashing. Our inspiration was to build a smart companion that is simple to set up but still provides real sustainability insights right when you click on the products. That's why we came up with EcoShop.

What it does

EcoShop is a smart browser extension that acts as your personal sustainability analyst while you shop online. When you browse a product on Shopee, EcoShop gets to work:

Analyzes Products in Real-Time: It sends product information to our powerful backend, which uses Google's Gemini 2.5 Flash LLM to perform a holistic sustainability analysis.
Grounded in Facts: We've enabled the LLM with Google Search capability. This means it can go beyond the product description to research a brand's labor practices, verify material claims, and uncover its true environmental reputation.
Clear, Personalized Scores: The deep analysis is distilled into a simple 0-10 score and a breakdown covering Material Composition, Production & Brand Reputation, and Circularity. Additionally, users can adjust the scoring weights to reflect what matters most to them.

How we built it

We designed EcoShop with a modern, decoupled architecture to ensure it was both powerful and scalable.

Backend: We built a modular API using Flask and Python. The logic is cleanly separated into an analyzer (for LLM interaction), a scorer (for calculations), and a central processor to orchestrate the workflow.
Database: We chose MongoDB as our intelligent cache. The results of expensive LLM analyses are stored once, allowing for instant, on-the-fly score recalculations for previously seen products, saving time and API costs.
Artificial Intelligence: We integrated the Google Gemini 2.5 Flash model. Our key innovation was using Function Calling to give the model two "tools": a google_search tool for research, and a submit_analysis tool. By forcing the model to call the submission tool, we guarantee a perfectly structured JSON output every single time, solving a major reliability challenge.
Frontend: The browser extension was built with lightweight, vanilla JavaScript, HTML, and CSS. Its only jobs are to scrape the page text and display the final analysis sent from our backend, keeping it fast and responsive.

Challenges we ran into

As first years students without much real world coding experiences, the reality of being thrust into such a competitive atmosphere where we had to learn and come up with a product in such breakneck speeds was nothing to sneeze at. Our combined experience does not amount to much but we still resolved to do our best to come up with our product EcoShop. Like Tony Stark trying to build something in a cave, with a box of scraps but the scraps are in fact our limited brainpower. I never thought that I would say "(git) PUSH" since my wife was giving birth, and I don't even have a wife, but this project made me live through a better experience.

Accomplishments that we're proud of

True Grounded Analysis: We gave our LLM the ability to use Google Search which means our analysis is dynamic and can uncover information that isn't on the product page, making our scores more accurate and trustworthy.
A Highly Efficient Caching System: By storing the detailed analysis in MongoDB and recalculating personalized scores on-the-fly, we created a system that is incredibly fast for returning users without re-running expensive API calls.
A Fully Decoupled, Professional Architecture: We successfully built and tested a system with a separate frontend, a modular backend API, a database, and an external AI service. This is similar to real-world architecture that's built to scale.

What we learned

We learned that we had to adapt to the data. We initially designed a scoring model with five rigid categories, but real-world product pages were too messy. We learned to pivot our design to a more holistic "Product Lifecycle" model that was more resilient to sparse data. Additionally, we learned that when working with LLMs, prompt engineering is not sufficient. We needed to understand how to use advanced features like Function Calling and Tool Use which is the key to building AI applications that is working with grounded data.

What's next for EcoShop

EcoShop is a powerful foundation, and we're excited about its future. Our next steps are:

Platform Expansion: Refactor the backend to create a "processor router," allowing us to easily plug in new modules for other major platforms like Amazon, Lazada, and Taobao.
User Impact Dashboard: Develop a web dashboard where users can see their shopping history, track their cumulative sustainability impact over time, and see how their choices create positive change.
Image-Based Analysis: Enhance our Gemini integration to process product images, allowing it to automatically detect plastic packaging, assess material quality from visuals, and identify other design-related sustainability factors.

Link to test our project!