A mobile app that turns every ra

PromptCube 初级 1小时前 115 浏览 6 点赞 约 2 分钟

I was pulling my hair out last month when I needed that one photo of the electrical conduit behind my kitchen wall. It was buried alongside family portraits and sunset shots, and the phone’s “mountains” and “documents” tags offered zero help. The native gallery just can’t make sense of a picture that has no text and lives outside the usual categories.

That frustration sparked Keeplea: a place where every “utility” snap—product packaging, government notices, wiring diagrams, even a fleeting screenshot—automatically lands and stays organized without any manual sorting. The app leans on semantic search so you can hunt down the exact moment you captured a circuit’s layout without scrolling through a chaotic feed.

What sets Keeplea apart is the backend mix. I built the UI with Kotlin Multi‑Platform (KMP) to share code across Android and iOS, leaned on Firebase for authentication and cloud storage, and tapped Gemini’s vision model to understand the content of each photo. Firestore’s vector search indexes the embeddings, turning vague mental cues into precise matches.

Right now the app is still in the “hard” phase. There’s no user base yet—search volume for this niche is almost nil, and most people don’t stumble upon such a problem in a store. I’m iterating on the categorization logic, polishing the onboarding flow, and hoping early adopters will give it a spin. If you’ve ever lost a critical snapshot in a sea of vacation photos, give it a look and let me know what works, what doesn’t, and what you’d add.

Website:

https://keeplea.app
Get the app:
https://get.keeplea.app

Keeplea wasn’t a weekend “vibecode” sprint. The idea simmered for weeks, then months, while I sketched architecture diagrams, ran experiments with different embedding models, and refined the UI flow. The term “vibecoding” feels cheap when it glosses over the countless late nights spent debugging vector queries and juggling native performance constraints. Using AI agents during development certainly helped, but the result is the product of systematic work, not a quick hack.

If you’re curious about building a mobile app that actually remembers the things you need later, check out Keeplea and share your thoughts.

GeminiKeepleaKMPFirestore

全部回复 (4)

咖啡续命折腾党 中级 1小时前
这种隐私问题确实让人心里发毛,要是真被传到什么公开库里就全完了。不过现在的厂商一般都会说走端侧处理或者加密传输,真想知道他们后台到底是怎么做脱敏的。
0 回复
全栈小李 高级 1小时前
昨晚熬夜等首发进去测了,结果排队排到怀疑人生,希望服务器这次能撑住别崩。
0 回复
内卷王调参侠 中级 1小时前
Good point on the PDF angle—those docs often contain way more sensitive stuff than people realize, bank statements, contracts, medical info. Even without bad intent, any external processing means your data leaves your device, and that's where I'd want clarity.
0 回复
深漂独立开发者 中级 1小时前
@内卷王调参侠 没错,光是 PDF 就能泄露那么多,更别说还有很多人不知道云处理其实已经帮它们打上了脚印
0 回复

发表回复

支持 Markdown 格式