gollum — hunting value-for-money at auction

gollum started with a real problem. I wanted to buy a particular kind of necklace, and the online auctions I was watching made it maddeningly hard to answer one question: which listing is actually the best deal? The cheapest item is rarely the best value, and comparing dozens of listings by hand — each with different condition, weight, materials and photos — doesn’t scale past the first evening.

So I built a tool that watches the auctions for me and ranks them the way I’d want to: by quality-for-price, not by price alone.

Watching the auctions

Under the hood, gollum runs a Playwright scraping pipeline. Auction sites are JavaScript-heavy and change often, so a headless browser that actually renders the page turned out to be far more robust than trying to reverse-engineer their APIs. It tracks the listings I care about, follows them as bids move, and keeps a tidy record of what’s out there right now.

Ranking value, not price

The interesting part isn’t the scraping — it’s the ranking. A cheap listing with poor condition and vague photos shouldn’t beat a slightly pricier one that’s clearly a better object. gollum scores listings so the best value floats to the top, which is exactly the judgement I was making by hand, just faster and without the fatigue.

To go further, I added on-demand LLM vision: point it at a listing and it reads the photos — condition, materials, details the text description leaves out — and folds that into the picture. It’s opt-in per listing, so it only spends effort where I’m genuinely interested.

Telling me when it matters

A tracker is only useful if it reaches you at the right moment. gollum sends alerts through an email service I host myself — the same infrastructure story I’ll write about separately. When something worth my attention appears, it lands in my inbox instead of me refreshing a page.

It’s a small, sharp tool built for one job, and it does that job well. That’s the kind of software I like: born from a real need, and genuinely used.