EcoLens - See the True Cost of What You Buy

Inspiration

90% of consumers want to shop sustainably. Almost none do it consistently not because they don't care, but because they have less environmental information at the moment of purchase. Amazon is the world's largest marketplace and shows price, reviews, delivery speed. Zero CO₂. Intangible packaging impact. That gap felt like an obvious, solvable problem. We built the missing layer.

What It Does

EcoLens is an AI-powered Chrome extension that activates on every Amazon product page and delivers a real-time sustainability verdict in under 5 seconds, no tab switching, no manual research.

Sustainability Score Card : 0–100 score with leaf rating, broken into Carbon Footprint, Packaging Waste & Brand Ethics sub-scores, injected directly on the product page
Greener Alternatives : 2–3 eco-friendlier Amazon swaps surfaced inline, ranked by score
Voice Briefing : tap once, hear the full sustainability summary read aloud via ElevenLabs
Cart Scanner : aggregates environmental impact of your full cart before checkout, with swap suggestions
Add-to-Cart Intercept : gentle friction nudge when a low-scoring product is added
Companion Dashboard : tracks cumulative CO₂ saved, purchase history & shareable monthly impact card

How We Built It

Stack: Plasmo (React + TypeScript, Manifest V3) · Next.js 14 · shadcn/ui · FastAPI (Python) · Supabase (PostgreSQL) · Google Gemini 2.0 Flash · ElevenLabs · Auth0 · Railway · Vercel

The core is an async multi-agent framework: ProductParser → (DatabaseLookup ∥ GeminiResearch) → ScoreAggregator → RecommendationEngine → VoiceNarrator. DOM scraping extracts product name, brand, ASIN, materials, and origin from public page data no Amazon API, no partnership needed. Every score is grounded against real open datasets: EPA eGRID, Open Food Facts, and GoodOnYou brand ratings to eliminate hallucination.

Challenges We Ran Into

Auth0 PKCE in a Chrome Extension : the callback URL format (https://<extension-id>.chromiumapp.org/) is non-standard; client ID is baked in at build time, so shipping the dev build breaks auth entirely
Parallel agent coordination : running DatabaseLookup and GeminiResearch in parallel without race conditions on the ScoreAggregator required careful async design in FastAPI
Grounding AI scores : an ungrounded sustainability score is worse than useless; anchoring Gemini to real open datasets was the key to trustworthy output
Shadow DOM isolation : injecting UI into Amazon's dynamic layout without breaking their CSS required fully scoped styles and shadow DOM boundaries

Accomplishments That We're Proud Of

Production-grade multi-agent framework, in 48 hours : five specialized agents, two running in parallel, built and deployed in a weekend. It's an architecture, not a demo
Under 5 seconds end-to-end : from page load to full score including a live Gemini call and DB lookup. Speed was non-negotiable
Works on every Amazon product, zero partnership required : any ASIN, any category, scored entirely from public page data
Multimodal + voice : Gemini reads product images and text together; ElevenLabs narrates the result. Accessible, hands-free sustainability intelligence
Real impact at scale : at 1M users, nudging just 5% of purchases greener = ~43,200 tonnes CO₂ avoided/year, equivalent to removing 9,400 cars from the road

What We Learned

Orchestration is the product. Anyone can call Gemini. The value is in how you structure agent handoffs, what you parallelize, and how you synthesize multiple signals into a score a user trusts in 3 seconds.

Open data changes everything. We trained zero models. EPA eGRID, Open Food Facts, GoodOnYou the data exists. The real work was connecting it cleanly to eliminate hallucination risk.

Friction can be a feature. The add-to-cart intercept was our most debated decision. What we learned: gentle, well-designed friction at the right moment changes behavior. The difference is in the copy, the design, and the timing.

What's Next for EcoLens

We want to keep building beyond the hackathon expanding to an Order History CO₂ Audit (retroactively score past Amazon purchases), weekly personalized impact email reports, and a community aggregation layer showing real-time collective CO₂ avoided by all EcoLens users. The multi-agent architecture is marketplace-agnostic Walmart, Target, eBay are next. Long-term: EcoLens becomes the sustainability layer of the internet, bringing environmental transparency to every purchase on every platform.