Intent Cart

Your Cart Has a Carbon Footprint

Every return costs more than money; it costs the planet. Over 5 billion pounds of returned goods end up in landfills each year, and reverse logistics adds about 15 lbs of CO₂ per return. Most of this waste comes from avoidable purchases.

Intent Cart helps stop waste before checkout.

What It Does

Intent Cart is a Chrome extension that reviews each product before you buy it. It reads live product data, compares against your real purchase history, and uses Gemini AI to flag return risk, accidental duplicates, and size mismatches in real time.

It also shows your environmental impact directly on Amazon and updates as you switch sizes, colors, and variants.

How We Built It

We built Intent Cart as a Manifest V3 Chrome extension with a Node.js/SQLite backend. A background crawler collects Amazon order history and builds a profile of your sizes, brands, and habits.

That profile is passed into Gemini 2.5 Flash with live product context. Gemini is the core reasoning layer, with no heuristic scoring. Everything is local-first: no data leaves your machine except the model API call.

Challenges We Ran Into

Amazon’s DOM changes constantly, so selectors break across categories. We solved this with layered fallback selectors and parsing strategies.

Manifest V3 service workers also terminate early, which disrupted async calls. We stabilized request flow and tuned prompts heavily to get reliable structured JSON from Gemini.

Accomplishments We’re Proud Of

Real-time risk analysis in ~800ms. Big win: Gemini catches subtle duplicate patterns (for example, near-identical products purchased months apart) that rules-based systems miss.

What We Learned

Gemini 2.5 Flash is ideal for browser UX: fast responses and stable JSON. Users respond more strongly to climate impact than savings.

Estimated avoided impact per prevented return on average: 15lbs of CO₂

ΔCO2≈15×(returns prevented)

What’s Next for Intent Cart Checkout interception modal for high-risk purchases Walmart, Target, and eBay support Crowdsourced return-risk data Return tracking feedback loop to improve recommendations