EcoTrack

dashboard1
dashboard2
chrome_ext
telebot

Inspiration

Our team was inspired by the growing frustration around greenwashing — brands making vague or misleading sustainability claims with little accountability. We wanted to build a tool that empowers people to shop ethically by making product sustainability transparent, accessible, and trustworthy. The idea of a Wikipedia of product lifecycles, that grows with community input, formed the heart of our solution.

What We Learnt

How to build and integrate AI pipelines for semantic product matching and summarisation.
The importance of entity disambiguation when dealing with messy product data across sites.
How to maintain performance when summarising thousands of lifecycle events.
How to design feedback loops between users and data to improve system intelligence over time.

How We Built It

EcoTrack consists of three main interfaces:

Chrome Extension: Displays sustainability scores in real-time as users browse products.
Dashboard Website: An interactive explorer that shows product lifecycles, semantic links, and personalised insights.
Telegram Bot: Lets users report product issues and receive eco-alerts with just a photo.

Under the hood:

OpenAI API: Used for vector embeddings and lifecycle summarisation
Postgres: Used for structured product and event storage (and also vector storage)
Tree-based summarisation algorithm: Minimises token use by summarising in chunks and recomputing only what's changed

Challenges We Faced

Entity Disambiguation: Products often appear under different names and URLs—we had to use vector similarity and metadata matching to group them correctly
Summarising Lifecycles: Storing and summarising hundreds of events per product quickly bloats LLM input limits. We built a custom tree-based method to chunk and compress history
Data Flow Design: Coordinating updates between the extension, bot, and dashboard required careful planning to avoid inconsistencies