Two things live here:
๐ฑ Sorted โ a free iOS app to track what you take, log doses, spot overlaps, and see what's running low. Calm, not clinical.
๐ ๏ธ Build Lab โ $0.99: exactly how I shipped it as a non-developer with AI.
Sprout keeps me honest.
everyone's posting "how to prompt claude code" today. the prompting was never my problem. i'd ask for a screen, get exactly what i asked for, and realize only then that what i asked for was wrong. did that maybe forty times before it shipped.
@BigMoneyBean the part that always wrecks my "tight routine" plans is the ones at different times โ morning's easy, the 2pm one is where it falls apart every time. same for you, or you got a system?
@rubenhassid this works till you hit something with no web stack. shipped an iOS app to the app store this way and the prompting was never the wall โ the wall was i couldn't describe what i wanted till the AI built the wrong version twice and i finally saw it.
@anshuc the visuals being all AI-generated is the part that'd scare me on a tracker โ easy to land in "generic health app" territory. how many passes did it take to get them feeling like yours and not stock?
@arthuryuzbashew what'd you change for the ASO pass โ subtitle, keywords, screenshots, all of it? about to do my own and the "hope so" is exactly where i'm at.
@takehisa_tech oh that reframe actually got me. "26 weekends" lands so much harder than "180 days" and it's the same number. what made you land on weekends specifically and not months?
@VladoIvankovic congrats on shipping. the "built for two" bit is the interesting call here. most trackers are a solo thing. how do you do the partner view without it tipping into feeling like surveillance?
@anshuc the training-data point is the part that actually lands for me. way more uikit out there for an agent to pull from. did swiftui fall apart everywhere, or mostly on the custom gesture/transition stuff? i'd expect the plain list and form screens to hold up fine.
the feature i use most is the most boring one: it tells me i'm three days from running out of something before i'm actually standing there with an empty bottle. nothing clever. i'd just genuinely miss it.
@henricreates@thoughts_app_ what's giving you the overdone feeling โ too many logos, or the proof crowding out what the app actually does in shot one? that's the only screenshot most people see before they swipe past, so it's the one worth sweating.
@beka_saparbek wait, you're training your own models now? for the offline gen or fine-tuning something existing? curious what the rtx6000 actually unlocks day to day.
@VaibhavSisinty been running a hand-rolled version of this for months, just markdown files the agent reads before it touches code. the part i can't tell yet: when an agent maintains its own wiki, does it stay accurate or quietly drift?
@JackdeS11 42 tok/s on-device for the 12B is honestly usable. have you pushed it on anything real yet or is it still demo-prompt territory? wondering where it starts to fall over.
half my timeline today is rediscovering that an AI works best off plain markdown it can read and maintain. i've run a scrappier version the whole time. every feature starts as a written brief before the model touches code. here's the path from a one-line idea to that brief.