My entire homelab is eBay: the servers, the Brocade switches (ICX6450 and ICX6430), the Ruckus APs, the old APC UPS (running on a LiFePO4 pack I swapped in when the lead-acid died), even the patch cables came as a lot of 20. The only new machine I own is my desktop. So I buy a lot of used RDIMMs, and I got tired of eBay search: kit notation ("32GB (2x16GB)" is not 32GB sticks), five different ways to write 2666 MT/s, lots, auctions, and "for parts" listings all mixed together.
DIMMsum pulls a matrix of ~51 specific searches twice a day, runs the titles through an LLM normalizer (DeepSeek, about $0.09 per 1,000 titles), and charts $/GB with median lines per speed grade. As of this week it also harvests eBay's sold/completed listings weekly, sothere's real sold-price history behind it with monthly market reports coming.
Monetization disclosure up front: outbound links are eBay Partner Network affiliate links. eBay pays. The site is free and there are no ads or accounts.
DIMMsum pulls a matrix of ~51 specific searches twice a day, runs the titles through an LLM normalizer (DeepSeek, about $0.09 per 1,000 titles), and charts $/GB with median lines per speed grade. As of this week it also harvests eBay's sold/completed listings weekly, sothere's real sold-price history behind it with monthly market reports coming.
Monetization disclosure up front: outbound links are eBay Partner Network affiliate links. eBay pays. The site is free and there are no ads or accounts.
Write-up of the build (why an LLM beats regex for eBay listing titles, and what parsing 18,000 of them costs): https://austinsnerdythings.com/2026/07/05/building-dimmsum-u...
Happy to answer anything about the approach or the data.