JBay

Inspiration

Buying and selling used electronics can feel like a guessing game. Buyers may not know whether a listing is accurate or fairly priced, while sellers have to describe their items and negotiate without much guidance. We built JBay to make those decisions clearer and secondhand transactions more transparent.

What it does

JBay is an AI-assisted marketplace for secondhand electronics. Sellers can upload photos and get help checking listing details, identifying visible defects, and describing an item. Buyers can browse listings, compare prices against recent sold listings, and negotiate with the seller through a shared deal session.

How we built it

We built the frontend with Next.js, 3.JS, and React, and the backend with Flask and Python. SQLite stores marketplace listings, photos, price observations, and negotiation sessions. Auth0 handles sign-in.

For listing assistance, a vision model analyzes product photos. OCR and validation rules help check model details and keep uncertain claims from being presented as verified facts. The pricing tool uses recent sold listings to estimate an average price and reports when there isn’t enough data. Negotiation uses a deterministic alternating-offer system, with separate private limits for the buyer and seller.

Challenges we ran into

A major challenge was making image-based listing assistance useful without letting it overstate what a photo can prove. A product’s exact model, functionality, authenticity, or condition may be unclear, so we added evidence checks, uncertainty labels, and seller confirmation.

Pricing was another challenge: comparisons are only helpful when they match the item and condition closely, and some products have very few recent sales. We surface data limitations rather than treating every estimate as equally reliable.

Accomplishments that we're proud of

We brought listing assistance, price research, and negotiation together in one marketplace flow. We’re especially proud of the safeguards around photo-based claims and the negotiation design: each side’s private price limit stays private, while both can participate in the same persistent deal session.

What we learned

Although AI can do many things, it works best as a decision-support tool: helping people gather evidence and explore options while leaving the final judgment to them.

What's next for JBay

Next, we’d like to expand the sold-price dataset, improve support for more product categories, and make listing reviews easier for sellers to correct. We would also like to incorporate 3D rendering of the product in the future.