Pricer

• 

  Page 1

• 

  Page 2

## Inspiration

Pricer came from a simple problem: clothes were piling up in my small NYC closet. I enjoy the hunt of finding clothing, but I do not enjoy the process of offloading it once I’m done with it. Listing items for resale takes time, research, and a lot of repetitive work, so I wanted to automate that workflow.

## What it does

Pricer takes a photo of an item and turns it into a market-based eBay listing. It identifies the item, researches real resale prices on the web, generates a title and description, and lets me track the item after it goes live so I can monitor price drift over time.

## How we built it

I built Pricer end to end by mapping the process I normally do manually and turning each step into an agent workflow. That meant going from image-based item identification to web research on comparable listings, then generating the listing copy and storing the item in ClickHouse for tracking.

The app combines a FastAPI backend, a React frontend, Gemini vision for identification, Nimble for live web search, x402 for paid analysis, and CDP wallet support in the frontend.

## Challenges we ran into

The biggest challenge was making sure the item was searched correctly. If the agent misunderstood the item, the pricing research could go in the wrong direction. I had to tweak the websites being searched, refine the query logic, and improve the agent flow so the search inputs matched the item more reliably.

## Accomplishments that we're proud of

I’m proud that the project actually does real work on the open web instead of just producing a mock demo. It identifies items, pulls in live market data, generates useful listing copy, and supports ongoing tracking after the item is listed. I’m also proud that the payment flow is integrated into the analysis path, so the system is not just informational but operational.

## What we learned

I learned how much agent quality depends on the structure around the model, not just the model itself. Good results required better source selection, better search queries, and a carefully ordered workflow. I also learned more about using web search as a pricing signal, rather than treating the model’s guess as the final answer.

## What’s next for Pricer

Next, I want to connect Pricer to the eBay Developer API so it can handle selling and price updates more directly. That would let the agent do even more of the workflow automatically instead of stopping at listing generation and tracking. I wasn’t able to build that part yet because eBay requires manual account approval before the API can be used, so it’s the main piece I’d add once access is approved.

Built With

• coinbase-cdp
• fastapi
• google-adk
• google-gemini
• nimble
• python
• react
• typescript
• x402