What was wrong
I buy almost everything secondhand, and the good stuff is buried. You scroll for an hour, find one thing, and then have to talk the seller down without being rude about it.
What I made
You talk to it. It asks about style, price and condition until it knows what you actually want, then a team of agents goes and searches.
When it finds the listing, there's a button: Ready to Lowball. It reads the photos and the listing and writes the message for you, pointing at the scuff on the sleeve and naming a number.
LowBall · exported from n8n · 15 nodes · scroll sideways →
The actual workflow, exported from n8n. A webhook in, an agent that writes the eBay search URL, Firecrawl scraping the listings, a second agent that determines the LowBall pick, a row saved to Supabase, and the answer sent back.
The hard part
This was before the models had voice built in. Wiring up ElevenLabs and getting the interface to animate cleanly while it listened took more work than everything else combined.
How it went
It won first place and $1,000, and over 100 people used it. It's retired now. It ran on n8n and I cancelled the subscription.
How it works
- Tell it what you're after. Say, a Carhartt jacket.
- It asks what a picky shopper would ask: style, price, condition.
- A team of agents searches the web and brings back the listing that fits best.
- Hit Ready to Lowball and it reads the photos and the listing.
- It writes the offer: the left sleeve is a little scuffed, so best I could do is $75. You send it.
What came out of it
- First place, $1,000, in a hackathon inside Nate Herk's AI community
- Built in 30 days with no coding
- 100+ users
Status · as-is
Agent retired. I cancelled my n8n subscription.
“It was very bad at first. But over time, it got better.”