Template

Quick Start

# Basic template-based optimization
uv run llmn optimize --pattern pescatarian --template --json

# With multiple patterns
uv run llmn optimize --pattern pescatarian --pattern slow_carb --template --json

# With reproducible seed
uv run llmn optimize --pattern pescatarian --template --seed 42 --json

# With goal-based targets
uv run llmn optimize --pattern pescatarian --template --goal "fat_loss:185lbs:165lbs" --json

Available Patterns

Combine patterns: --pattern pescatarian --pattern slow_carb

Workflow

Step 1: Identify User's Diet Type

Map to patterns:

• "I'm pescatarian" → --pattern pescatarian
• "I'm vegetarian" → --pattern vegetarian
• "I'm vegan" → --pattern vegan
• "I eat everything" → --pattern mediterranean or --pattern paleo

Step 2: Identify Diet Style

• "Slow carb" → add --pattern slow_carb
• "Keto" or "low carb" → use --pattern keto
• "Mediterranean" → use --pattern mediterranean

Step 3: Identify Goals

Use the --goal flag:

• Weight loss: --goal "fat_loss:185lbs:165lbs" (current:target)
• Maintenance: --goal "maintenance:165lbs"
• Muscle gain: --goal "muscle_gain:165lbs:175lbs"

Step 4: Run Optimization

uv run llmn optimize --pattern pescatarian --pattern slow_carb --template --goal "fat_loss:185lbs:165lbs" --json

Step 5: Present Results

Template output has per-meal structure:

{
  "meals": {
    "breakfast": {
      "foods": [
        {"description": "Egg, whole, cooked, fried", "grams": 150, "slot": "protein"},
        {"description": "Edamame, frozen, prepared", "grams": 100, "slot": "legume"},
        {"description": "Kale, raw", "grams": 100, "slot": "vegetable"}
      ],
      "totals": {"calories": 450, "protein": 35}
    },
    "lunch": {...},
    "dinner": {...},
    "snack": {...}
  },
  "daily_totals": {"calories": 1920, "protein": 175}
}

Present each meal with its foods and macros.

How It Works

Phase 1 - Selection (Discrete):

1. For each meal, select one food per slot (protein, legume, vegetables)
2. Enforce diversity: once a food is selected, exclude it from future meals
3. Selection is random with optional --seed for reproducibility

Phase 2 - Optimization (Continuous):

1. Run small QP (~16-20 variables) with only selected foods
2. Use slot-specific target portions (protein=175g, legume=125g, vegetables=100g)
3. Optimize to meet calorie/protein constraints

Tips

1. Use --seed for reproducibility: Same seed = same meal plan
2. Run multiple times: Different random selections give variety
3. Combine patterns: --pattern pescatarian --pattern slow_carb is powerful
4. Check the slots: Each food shows which slot it fills (protein/legume/vegetable)

Comparison with Other Modes

Example Session

User: "I want realistic meals, not the same vegetables everywhere"

1. Ask: Diet type? (pescatarian/vegetarian/vegan/omnivore)
2. Ask: Any diet style? (slow-carb/keto/mediterranean)
3. Ask: Goal? (weight loss/maintenance/muscle gain)
4. Run:

   uv run llmn optimize --pattern pescatarian --pattern slow_carb --template --goal "fat_loss:185lbs:165lbs" --json
   

5. Show results:
   • Breakfast: Eggs 150g, Edamame 100g, Kale 100g
   • Lunch: Cod 235g, Kidney beans 216g, Zucchini 161g
   • Dinner: Salmon 170g, Lentils 113g, Carrots 149g
   • Snack: Peanuts 30g
6. Offer: Different seed for variety, adjust goal, generate recipes