OffloadAI App

Indie iOS dev. Building OffloadAI — 100% local & private photo manager powered by Apple Intelligence.

🛡️ Zero cloud risk
🏗️ #BuildInPublic #iOSDev

[OffloadAI App @offloadai_app]

@mitansh_j07 Thanks! Bridging the gap between rigid administrative boundaries ("Bay Lake") and how human memories actually work ("Disney World") is easily one of the most rewarding puzzles in this build. ✨

[OffloadAI App @offloadai_app]

Reverse geocoding has zero human empathy.
Apple Maps says Disney World is in "Bay Lake, FL" —Nobody says that!
Turning 2D GPS points into a human hierarchy (Orlando ➔ Disney World ➔ Hollywood Studios ➔ Attraction) is absolute hell. 🤯📐
#buildinpublic #iOSDev

[OffloadAI App @offloadai_app]

@devoxisfig Definitely spatiotemporal edge cases: photos taken at 2 AM across timezones, and assets missing GPS metadata right in the middle of a trip. Getting the timeline to stay intuitive without losing a single file took the most iterations. 📐

[OffloadAI App @offloadai_app]

Most apps test with 10 fake mock photos and call it a day.

Before TestFlight, I put OffloadAI through 78 hardcore tests against a real library of 50,000+ assets.

Zero shortcuts. Here is what we actually tested 👇

#buildinpublic #iosdev

[OffloadAI App @offloadai_app]

@zeal_m_ Real-world edge cases broke it immediately—specifically midnight timezone cutoffs (like shooting past 1 AM on vacation) and dirty map metadata. Mocks are polite; real photo libraries are pure chaos. ��

[OffloadAI App @offloadai_app]

@Ariella1213 Sharp catch! Count-only checks are a classic trap. I actually assert full Set equality on unique asset identifiers before and after recovery—zero missing, zero silent duplicates. Precision is non-negotiable. 🛡️

[OffloadAI App @offloadai_app]

@iamvishal16_ios The Simulator is a liar! Mac silicon hides all the real thermal throttling and disk I/O bottlenecks. That’s exactly why I stopped trusting mock runs and audited directly on a physical iPhone. 📱🔥

[OffloadAI App @offloadai_app]

Result: 78/78 passing. 100% on-device. Zero cloud.

Pretty UI is just the surface; bulletproof engineering is how you respect user memories.

TestFlight invites rolling out soon. 🛡️

#buildinpublic

[OffloadAI App @offloadai_app]

1/ Real disk I/O & power-loss recovery
2/ Surviving prompt injection & sandbox jailbreaks
3/ Crossing timezones & 4AM date cutoffs without losing 1 photo
4/ Zero-overlap geometry at scale
Real data exposes bugs that mocks hide.
#iosdev

[OffloadAI App @offloadai_app]

戰果：78 項測試全部通過（100% Pass）。
100% 本機運行、零雲端。

靚 UI 只係包裝，做到金剛不壞先係對用家回憶嘅基本尊重。 TestFlight 倒數中。🛡️
#buildinpublic

[OffloadAI App @offloadai_app]

多數相片 App 測試只用十幾張假相做樣。

出 TestFlight 之前，我直接用真機 5 萬張真實相片，同 OffloadAI 跑咗 78 項硬核測試。

唔玩虛假 Mock，睇下我到底測咗啲乜 👇
#buildinpublic #iosdev

[OffloadAI App @offloadai_app]

1/ 實體磁碟讀寫與斷線回滾保護
2/ 擋住惡意 Prompt Injection 同沙盒越權
3/ 跨時區與凌晨 4 點換日線「零漏相」守恆
4/ 大規模渲染零重疊幾何驗證
只有真數據先逼得出隱藏 Bug。
#iosdev

[OffloadAI App @offloadai_app]

Result: 78/78 passing. 100% on-device. Zero cloud.

Pretty UI is just the surface; bulletproof engineering is how you respect user memories.

TestFlight invites rolling out soon. 🛡️

#buildinpublic

[OffloadAI App @offloadai_app]

1/ Real disk I/O & power-loss recovery
2/ Surviving prompt injection & sandbox jailbreaks
3/ Crossing timezones & 4AM date cutoffs without losing 1 photo
4/ Zero-overlap geometry at scale
Real data exposes bugs that mocks hide.
#iosdev

[OffloadAI App @offloadai_app]

零雲端、唔睇相片私隱嘅前提下，OffloadAI 完全用「物理運動特徵」去判斷：
相片坐標嘅空間散佈範圍
時空聚類嘅密度變化
只要你真��行過，物理足跡係呃唔到人嘅。
#buildinpublic #iOSDev

[OffloadAI App @offloadai_app]

搞相片自動分類，最怕遇到「同類但唔同命」嘅地方。

蠟像館行兩步影完，本質係單一室內展館； 但羅浮宮大到行幾萬步，仲連住成個花園。
如果寫死名單硬兜，未免太低能。點樣聰明處理？👇 #buildinpublic #iOSDev

[OffloadAI App @offloadai_app]

Instead of analyzing pixels, OffloadAI simply decodes how your footsteps felt:
Did you linger around a corner?
Or did you walk across miles?
Motion tells the real story.

#buildinpublic #iOSDev

[OffloadAI App @offloadai_app]

How do you group photos when places share the same category?

A wax museum is just a single building. The Louvre spans 700,000+ sq ft across wings and gardens.
Hardcoding names is lazy. So how do you solve it? 👇

#buildinpublic #iOSDev

[OffloadAI App @offloadai_app]

@Nanjialonguazt 暫時好像無法避免手動。

[OffloadAI App @offloadai_app]

地圖 API 永遠最缺乏「人性」。 做相片自動分類先發現，GPS 永遠死板板話迪士尼喺「Bay Lake, FL」。但全世界邊有人會同朋友講「我上星期去咗 Bay Lake」？
要將死板嘅 2D 座標，自動還原成人腦認知嘅（奧蘭多 ➔ 迪士尼世界 ➔ 影城 ➔ 景點），真係搞到頭都爆。🤯 📐
#buildinpublic #iOSDev