GUARDED AI MVP

Eat what you actually want.
Let CookCards balance the macros.

Start with a real dish — carbonara, wraps, bowls, pasta, oats. CookCards adjusts ingredient amounts using verified nutrition data while protecting the dish’s flavor identity.

Not another “AI recipe generator”

AI does not calculate macros

LLMs can hallucinate calories, protein, fiber, and servings. CookCards treats AI numbers as untrusted.

Database calculates

Nutrition is estimated from structured ingredient data, designed for USDA/FDC integration.

Rules validate

Ingredient sandbox, flavor templates, macro targets, and realistic cooking checks reduce recipe failures.

Why normal AI recipes break trust

Typical AI recipe

“This salad has 30g protein” — but the model guesses portions, forgets dressing calories, and treats vague ingredients like real nutrition math.

Failure mode: oil, sauces, nuts, cheese, and serving size quietly blow up the macros.

CookCards guarded recipe

AI can suggest flavor. But grams, calories, protein, fiber, and fat come from structured ingredient data and rule checks.

Rule: even 1 tbsp olive oil has to show up in the numbers.

Guarded generation engine

1. CravingUser says what they actually want.
2. TemplatePick a proven flavor template.
3. SandboxUse only real grocery ingredients.
4. VerifyCalculate macros from data.
5. WriteAI formats recipe after numbers are locked.
Principle: AI writes. Database calculates. Rules validate.

We don’t let AI invent taste

Flavor identity protection

Each CookCard starts from a curated dish template. Some ingredients are locked because they define the dish. Some are adjustable because they affect macros. Some swaps are forbidden because they would ruin the meal.

New principle: Humans define flavor. Algorithms balance macros. AI writes the card.

Example: Carbonara

Locked: pasta, egg/yolk texture, parmesan/pecorino, black pepper, salty cured-meat note.

Adjustable: pasta grams, egg white ratio, parmesan amount, pancetta/turkey bacon amount, Greek yogurt max amount.

Forbidden: beans, salsa, chia seeds, random chicken breast chunks.

Ingredient sandbox, packaged for real flavor

We restrict the early MVP to standard, high-quality groceries — then use professional flavor math and flavor templates to turn bland “gym bro” staples into CAVA, Chipotle, and comfort-food style meals.

Rebalance a dish to your macros

Fake-door test: production optimization engine is not connected yet.

Reading your macro targets...
Protecting dish identity...
Adjusting ingredient amounts...
Checking target vs actual macros...
Preparing beta access...

Mock result: Macro-balanced Carbonara

600 → 615calories, +15
45g → 47gprotein, +2g
60g → 58gcarbs, -2g
15g → 16gfat, +1g
What changed:
Reduced pasta from 100g to 75g, added egg whites for protein, kept parmesan at 15g for flavor, limited Greek yogurt so it stays creamy instead of fake.

Craving-first generator demo

Demo note: This is a fake-door validation test. The production USDA/FDC + LLM engine is not connected yet.

Matching your craving...
Choosing flavor template...
Building ingredient base from sandbox...
Verifying calories, protein, and fiber...
Creating cooking steps...

Trust promise

  • No AI-guessed macros
  • Nutrition estimated from structured ingredient data
  • Only grocery-store ingredients
  • Realistic cook time checks
  • Flavor template, not random ingredient soup

10 curated macro-balanced CookCard templates

Click any recipe. The full verified CookCard below updates with the same structured template.

Full verified CookCard

Validation questions

Would a macro rebalancer beat another meal prep app?

This MVP tests whether macro-conscious users want ingredient-level rebalancing for dishes they already crave — not another recipe picker.

Meal prep question: Would you rather generate one dinner, or 4–5 servings for Sunday meal prep?
Pricing question: Would $4.99/month, $19.99 lifetime early access, or free-only feel right?