What's for Dinner?
When meal planning is more than a recipe
Monday, 4:30 PM. My wife and I are standing in the kitchen, looking at each other. "What's for dinner?" We have this conversation three times a week. And three times a week it ends the same way: something with pasta. Or ordering pizza. Not because we can't cook. But because meal planning for our family is an optimization problem that no cookbook solves.
When food gets complicated
Our son is autistic. That doesn't mean he doesn't want to eat. It means textures play a role that no recipe book can imagine. Nothing mushy. Nothing that mixes on the plate. Familiar shapes, predictable consistencies. Then there are allergies — for us, several of the 14 EU main allergens. And then our daughter, who suddenly wants to be vegetarian. And my wife, who's trying low-carb.
Normal meal planners fail at the second family member. They don't know autism-friendly options. They don't understand that "potatoes" as a wish doesn't mean "potatoes every day" but once or twice as a side dish. They don't get that "casserole" means exactly once on the weekend — because that's the texture our son accepts on weekends, but not during the week.
An app that knows our family
So I built FamilienKoch. An app where every family member has their own profile. Role, allergies, dietary preferences, likes, dislikes — and when needed, autism-friendly options: familiar textures, no mixing on the plate, predictable meals.
The AI behind it doesn't make random suggestions. It interprets semantically: "Italian" means two to three Italian-style dishes across the week, not pasta every day. "Potatoes" means side dish, not main course. And if our son accepts fish sticks on Tuesdays, then fish sticks are on the plan for Tuesday — because predictability matters more than variety.
From plan to shopping list in seconds
The weekly plan automatically generates a shopping list. Sorted into 13 categories, assigned by family member — Dad shops at the Turkish specialty store, Mom at the regular supermarket. Which supermarket? The one with accessible entry, contactless payment, and the best opening hours. All via OpenStreetMap, no Google.
And when Monday at 4:30 PM nothing is planned: "What can I cook?" — enter the ingredients still in the fridge, and the AI makes a suggestion. One that works. For everyone.
9 days, 61,000 lines of code
FamilienKoch was built in 9 days. React Native, TypeScript, Claude as AI backend. 80% of the code was written by AI, 80% of the architecture decisions were mine. Vibe coding in its purest form: I say what I want, the AI builds it, I test and refine. 13 iterations per day.
The COCOMO model estimates traditional development effort at 18 months with a 10-person team. Cost: two million dollars. My costs: nine evenings after work and a Claude API bill I'd rather not talk about.
Why this matters
Not because of the technology. But because "What's for dinner?" is the question every family knows — but not every family has the same problem behind it. For most, it's convenience. For us, it's inclusion. Making sure our son can eat what's on the table. That nobody has an allergic reaction. That the week becomes plannable.
And yes — since then, Monday at 4:30 PM is more relaxed. We look at the plan instead of looking at each other.
— Philipp