Seven months later: what I built since March

The blog went quiet. The work did not.

My last post was in March. Since then I shipped less noise and more real things. Some projects moved a lot, some I put on hold on purpose. Here is where everything stands.

CookBook AI: redesigned around three needs

I rebuilt CookBook AI around what people actually do in a kitchen: import a recipe from a link, plan the week, get the shopping list. Everything else became secondary.

• Share a recipe from Safari or YouTube straight into the app
• "Fill my week" plans lunch and dinner in one tap, without duplicates and without allergens
• The shopping list follows the plan, rounds quantities to what you actually buy, and sorts everything by aisle

The design now follows Apple's guidelines (system font, Dynamic Type, real dark mode), and on-device Apple Intelligence handles imports even without an API key. It is used every day at home on an iPad mini. What is left before the stores is mostly paperwork: ads setup on Android, the premium product, signing.

ScolIA: a school app for iPad

The biggest new project. ScolIA is a native SwiftUI app for primary school: the French curriculum, the level of the best international schools, and future skills like money, code and AI, engineering and critical thinking.

The rule that shaped everything: the child never writes. Every exercise is touch-based (16 types), every lesson ends with a challenge that asks you to use what you learned somewhere else. Today: 117 lessons, about 1,400 exercises, 100% offline.

OmniDrop: stop, rethink, rebuild

I have to be honest about OmniDrop. The March version had a lot of features and not enough reliability. Meanwhile the market moved: AirDrop now talks to Quick Share on recent Android phones, and LocalSend keeps getting better.

So I reset the plan. The new core speaks the open LocalSend v2 protocol. Local transfer will be free and unlimited. The chat and the OmniDrop ID are on hold. The focus is what is still missing out there: a clipboard shared across every system, and sending files outside your Wi-Fi without an account. Nothing else ships before a transfer works 10 times out of 10 on two real devices.

New things I did not plan

• Bonne Étoile: a funny isekai text RPG for iPhone. Ten chapters, d20 dice, and an on-device AI game master that improvises when you type something the story did not expect.
• AION 2 Companion: a passive tool for the game AION 2. A transparent overlay draws the best route on the map, and an advisor reads my character from the official public page. No memory reading, no automation: the account stays safe.
• MotionStudio: videos written in code, rendered at 60 fps. The first ones present my own apps, filmed on the real builds.
• Plans3D: a prototype site where a house plan draws itself like an architect sketch, rises into a 3D model, then becomes a real-time render with a sun you can move.
• L'Écorce: yes, a novel. A heroic fantasy saga planned in five volumes. Volume 1 is drafted, 40 chapters, and waiting for my reading.

Behind the scenes: a better way to work with AI

I keep building with Claude, but the setup changed. My workspace now has guard rails that run on their own (a hook that blocks dangerous commands, a checkpoint every 75 actions), small specialised helpers (one that reads, one that verifies), and a command that proposes, tests, then keeps improvements to the setup itself. Fewer rules to remember, more checks that run automatically.

What I put on hold

Ephemera, BridgeGen, EarthPulse, GhostKitchen, NightOwl, PlateScore, Nova and StickerVibe AI are still on the site, now marked On hold. Their MVPs exist, but a solo dev cannot push 20 projects at once. I would rather finish a few.

What comes next

1. OmniDrop: reliable local transfer, tested on two real devices
2. CookBook AI and TipLog: store submission
3. More presentation videos, including one for this site

Thanks for reading. If one of these projects speaks to you, tell me: it helps me choose what to finish first.