EcoAware — Devpost Submission
Inspiration
Every day, millions of shoppers encounter vague environmental claims — "eco-friendly," "green," "natural" — with no way to verify them. The FTC has documented that most general environmental benefit claims are unsubstantiated. I wanted to build something that cuts through the noise and gives shoppers honest, grounded sustainability information at the exact moment they're deciding whether to buy.
What it does
EcoAware is a Chrome extension that automatically analyzes product sustainability when browsing Amazon. It scrapes the product listing — title, bullets, specs, badges, description, and buyer reviews — then scores sustainability across 8 weighted factors using a deterministic engine grounded in FTC Green Guides.
It detects greenwashing by matching claims against rules derived from federal environmental marketing standards. On top of the deterministic scoring, Claude AI provides a contextual verdict — checking whether reviews contradict the listing, whether claims make sense for the actual product, and summarizing what real buyers say about sustainability. ElevenLabs narrates the verdict aloud. For deeper questions, users can chat with a companion Sustainability AI Advisor built on Patriot AI.
How I built it
I started with a heuristic keyword-matching scorer and quickly realized it wasn't defensible — judges and users would ask "how did you get 78/100?" and there wouldn't be a good answer. So I rebuilt the engine from the ground up using structured JSON knowledge files derived from deep research into FTC Green Guides, eco-certification standards, and carbon claims policy.
The scoring engine evaluates 8 buckets — certifications, materials, packaging, durability, energy, transparency, ethics, and AI context — with weights that shift across 11 product categories. A toothbrush is scored differently from a laptop because sustainability means different things for different products.
I then layered Claude AI on top as an 8th scoring bucket at 10% weight — enough to nudge a borderline score but never enough to override the deterministic fundamentals. The AI also generates spoken-ready narration and review summaries. ElevenLabs provides text-to-speech so users can listen to the verdict hands-free.
The entire extension runs client-side with vanilla JavaScript — no frameworks, no build step, no backend.
Challenges I ran into
Greenwashing false positives were the biggest headache. The initial rule engine flagged a bamboo toothbrush for claiming "biodegradable" — but bamboo IS biodegradable. I had to build material-aware suppression lists so the engine understands that certain claims are factual for certain materials. Getting this right required going through the FTC Green Guides to understand when a claim is genuinely misleading versus when it's just stating a material property.
The "Unknown" problem. The confidence system was too conservative — it labeled almost every product without a major certification as "Unknown," which made the tool feel broken. A genuinely unsustainable plastic product and a genuinely good bamboo product both showed the same label. I had to rethink what "Unknown" should actually mean and limit it to cases where there truly isn't enough listing data to work with.
Balancing AI influence. I wanted AI to add value without dominating. If the AI hallucinated a high score, it shouldn't override real evidence. I solved this by capping AI at 10% weight — mathematically it can only swing the final score by ±5 points.
Accomplishments that I'm proud of
The material-aware suppression system is something I haven't seen in competing tools. Most greenwashing detectors use simple pattern matching and would flag "biodegradable bamboo" the same way they flag "biodegradable plastic." EcoAware doesn't.
The 8-bucket, 11-category weighted scoring architecture makes every score defensible. I can explain exactly why a product scored 56 — which buckets contributed, which signals were found, which were missing.
The three-tier tone system adjusts how the AI communicates based on greenwashing risk — firm and direct for high risk, constructive for moderate, encouraging for clean products.
What I learned
Building a "simple" scoring engine is harder than building an AI wrapper. The deterministic layer required more thought, more edge-case handling, and more research than the AI integration. Anyone can call an API — the hard part is making sure the data going into it is correct and the results are grounded.
I also learned that AI is best used as a layer on top of deterministic logic, not as a replacement for it. The deterministic engine provides consistency and explainability. The AI provides nuance and natural language. Together they're stronger than either alone.
What's next for EcoAware
• Expanding to more retailers beyond Amazon • Building a product comparison feature to evaluate alternatives side by side • Adding lifecycle impact estimates based on material composition • Accurately comparing carbon footprints between conventional and sustainable alternatives — showing the real environmental difference, not just a score • Building a thorough product classifier trained on labeled sustainability data to move beyond keyword heuristics toward genuine understanding of product environmental impact • Publishing to the Chrome Web Store for public use
Built with
• JavaScript (vanilla, no frameworks) • Chrome Extension Manifest V3 • Anthropic Claude API • ElevenLabs Text-to-Speech API • Patriot AI (GMU) • FTC Green Guides (16 CFR Part 260)
Try it out
• Extension: Load unpacked in Chrome from the project folder • Sustainability AI Advisor: Chat on Patriot AI
Log in or sign up for Devpost to join the conversation.