design.md

Design: RecipeAI

1. Introduction

A minimal end-to-end proof-of-concept pipeline using multi-agents concept:

1. Planning Agent: Generates meal outlines based on user constraints.
2. Chef Agent: Produces detailed recipe steps from the meal outline.
3. Nutrition Agent: Computes nutritional breakdown of the generated recipe.

This document covers high-level architecture, data flows, and implementation steps.

2. Goals and Constraints

Goals

• Validate multi-agent orchestration in LangChain.
• Produce a working pipeline: Planner → Chef → Nutrition.
• Enable iterative development and prompt tuning.

Constraints

• User-selected cuisine (e.g., Italian, Indian, Mexican).

• Maximum prep time (e.g., ≤30 minutes).

*Additional dietary restrictions and preferences (vegetarian, dairy-free, likes/dislikes) will be integrated in future iterations.

3. Architecture Overview

+---------+     +---------+     +-------------+
|  User   | →─▶ | Planner | →─▶ |   Chef      |
|  Input  |     | Agent   |     |   Agent     |
+---------+     +---------+     +-------------+
                                     |
                                     ↓
                               +-------------+
                               | Nutrition   |
                               | Agent       |
                               +-------------+
                                     |
                                     ↓
                               +-------------+
                               |  Output:    |
                               | Recipe +    |
                               | Nutrition   |
                               +-------------+

Agents communicate via JSON payloads validated by Pydantic models.

---

4. Component Details

4.1 Planning Agent

• Function: Interpret user constraints and propose a single meal outline.
• Prompt Structure: Few-shot examples plus JSON schema.
• Output Model:

from pydantic import BaseModel
from typing import List

class Ingredient(BaseModel):
    item: str
    qty: str  # e.g. "100g"

class Plan(BaseModel):
    meal: str
    ingredients: List[Ingredient]

4.2 Chef Agent

• Function: Expand meal name + ingredients into cooking steps.

• Prompt Structure: Input JSON Plan, request recipe fields.

• Output: Structured JSON or Markdown containing:

  • Title
  • Prep & cook times
  • Steps (numbered list)
  • Serving size

4.3 Nutrition Agent

• Function: Lookup macro & micro data per ingredient.

• Implementation Steps:

  1. Parse qty into grams or common units.
  2. Query USDA FoodData Central API (or Edamam).
  3. Aggregate calories, carbs, protein, fat, and key micros.
  4. Return JSON:

{
  "calories": 450,
  "macros": {"carbs": 60, "protein": 20, "fat": 10},
  "micros": {"iron_mg": 4.5, "vitamin_c_mg": 20}
}