– an aggregated search engine for Sweden’s used sales market
Background
My brother-in-law ran over my sunglasses with a string trimmer.
No worries, time to start hunting for used sunglasses. I’d already scoped out some nice frames, but they’re expensive. I was looking for Tripoint, either the Reschen, Treriksröset, or Lake Victoria model. A lot of brands look similar, but these ones fit me better.
Alright, time to log into each app and site and start searching: Facebook Marketplace, Blocket, Tradera, Vinted, Sellpy. If you find something at a decent price it’s usually worth the hassle, but even used ones were pricey to me. So I kept searching every few days until I started getting annoyed with the whole process.
I decided to try building something with AI that would make all of this a bit easier. I’m using Claude Code to build a foundation that searches at least Tradera and Blocket.
I’m singing hallelujah at how easy it is to get an initial prototype up and running that gets almost everything right. Even so, it’s soon time to shift focus; this project is becoming less about how to build a search engine and more about how I relate to AI.
Sayings
I’ve been having a feeling, a feeling I haven’t heard anyone yet put into words. I keep thinking about all these magical phrases in other languages that are hard to translate but make complete sense in their original language.
I sincerely hope someone finds the right words to describe my thoughts in the future, because I can’t be the only one feeling this.
I also realize the irony of trying to use a language model in my work and still not being able to find the right words to describe what’s going on.
Vorführeffekt (demonstration effect)
When you want to show somebody something you can do but it goes wrong because they’re watching you.
🇨🇳 報復性熬夜 (revenge bedtime procrastination)
A phenomenon in which people who don’t have much control over their daytime life refuse to sleep early in order to regain some sense of freedom during late night hours.
🇯🇵 口寂しい (Lonely Mouth) When you’re not hungry, but you eat because your mouth is lonely.
I feel like I have to be incredibly precise. Overly precise. And then some. I’ve spent years writing design documents and requirement specs, sitting in IT support, writing bug reports, coaching sports teams, raising kids.
I worry about everything, prepare for everything, always have a plan. But it’s still not enough. Because every single day I forget that AI is blind, can’t reason, and yet still manages to fool and impress me every day. Like the world’s best Nigerian prince.
At first, my thoughts are solely:
- Let’s search for something.
- Results appear.
- I get happy and think everything works.
And then you compare against the original sites, because it felt like more results should have trickled in the following day. Sure enough, if I go straight to Tradera or Blocket, I see a discrepancy in the number of hits.
So our conversation goes something like this (paraphrased):
Result
I’ve really kept pushing to build something I can both use and be proud of.
- Extra sources
- Sticky headers
- Sorting by price or date
- Settings
- Rearranging cards in the right-hand column
- Multiple users
- Toggling sources on and off
- Saved searches (click to reuse instead of retyping)
- Extra animation when clicking “search”
- Auto-scroll down to the search results
- Animated favicons to show status while waiting for results
- Hand-drawn logo
- Animated logo
- Android app shell
- Syncing settings and results per user
- Tons of layout changes
- Custom layout for night theme
- Watchlists that email you daily if the site finds new listings for you overnight
I’m happy with what I’ve built. I use the app almost exclusively to look for used stuff in Sweden (that can be shipped). I’m especially proud of the parts that were made without AI: the logo, the typography, and the sticky headers.
I have yet to buy any sunglasses.