Hi, I’m Sarah—software engineer in Vancouver—and in this cozy all-nighter devlog I’m working on my AI outfit generator app (aka my attempt to “try on” my closet without turning my room into a disaster). I start with a pretty ambitious to-do list: fix bugs, save generated outfit IDs so I don’t keep paying for repeated image generations, build a liked-outfits gallery, and then try to add an online shopping try-on flow. Along the way, the outfit transfer feature glitches so hard it literally gives me an evil twin behind me, which is… not the vibe, but also kind of hilarious. On the build side, I walk through how I’m using Supabase (tables + storage) to save generated outfits, then I add an “is_liked” boolean so I can filter a cute gallery of my favorites. I’m also reviewing PRs since the repo is public, and I’m using Graphite to keep reviews cleaner and stack my changes instead of dumping everything into one massive PR. Then we do a quick desk tour (lighting is genuinely the secret sauce), and I attempt web scraping for shopping try-ons—only to get blocked by sites like Aritzia and pivot to a screenshot-based approach. The biggest takeaway: half of coding is building, and the other half is figuring out why something that worked yesterday suddenly broke. In my case, it was the model (Gemini 2.5 flash image) getting disabled, and regenerating the API key saved my sanity.