CulinaX: A Pakistani Smart Cooking Assistant Fusing Rule Engines and Generative AI

CulinaX is a smart cooking assistant designed specifically for Pakistani cuisine, innovatively combining deterministic rule engines with generative AI to deliver a complete cooking experience from pantry management to AI chef conversations. The project is developed using the Flutter framework, with a backend based on Firebase, integrating Google Gemini and OpenAI's generative AI capabilities to create a hybrid intelligent system that balances precision and creativity.

Original Author and Source

• Original Author/Maintainer: tahanawab4848
• Source Platform: GitHub
• Release Date: May 27, 2026

Project Overview

CulinaX is positioned as a smart cooking assistant exclusive to Pakistani cuisine. Its core innovation lies in a layered architecture design that clearly separates deterministic rule engines from generative AI intelligence, ensuring both the reliability of core functions and a creative user experience.

Rule Engine Layer (Deterministic Layer)

Responsible for precise calculation and consistent functions:

• Pantry inventory management (recording ingredient storage time, shelf life, and expiration reminders)
• Smart ingredient matching (calculating the match between recipes and available ingredients)
• Recipe ranking (recommendations sorted by ingredient match degree)
• Automatic shopping list generation (listing items to buy when ingredients are missing)
• Meal plan scheduling

AI Intelligence Layer (Generative Layer)

Provides creative services via Gemini/OpenAI API:

• AI Chef chatbot (natural language interaction)
• Smart recipe generation (creating original recipes based on available ingredients)
• Explanation of recommendation reasons (illustrating the principles of ingredient pairing)
• Ingredient substitution suggestions
• Step-by-step cooking guidance (real-time prompts + voice broadcast)
• Smart meal suggestions (based on eating habits and pantry status)

Tech Stack

• Frontend: Flutter (Provider state management)
• Backend: Firebase (Authentication, Firestore, Storage)
• AI Integration: google_generative_ai package (Gemini) + http package (OpenAI as alternative)
• Rule Engine: services/recipe_engine.dart
• Device Features: Barcode scanning, image recognition (on-device)

Data Model

Core Firestore Collections:

• users (user information)
• pantry_items (ingredient list + shelf life)
• recipes (recipe database)
• grocery_list (shopping list)
• meal_plans (meal plans)
• ai_interactions (AI conversation history)

1. Login Initialization: After registration, add pantry items (e.g., chicken, rice) and set their shelf life
2. Smart Recommendations: The homepage displays AI-selected recipes and their recommendation reasons
3. AI Conversation: Use natural language to ask for cooking advice in the AI Chef interface
4. Recipe Generation: Generate original recipes based on available ingredients (suitable for scenarios like dorm dinners)
5. Recipe Details: View recommendation reasons and one-click ingredient substitution function
6. Cooking Mode: Step-by-step guidance + AI prompts + voice broadcast support

Advantages of Rule Engines

• Precise calculation and predictable results
• High efficiency in structured data processing
• Offline availability
• Good data privacy (local processing of pantry data)

Advantages of Generative AI

• Flexible natural language understanding
• Creative content generation (new recipes, substitution suggestions)
• Personalized explanations and conversations
• Continuous learning and improvement capabilities

Combining the two ensures both reliable core functions and a creative experience, which serves as a reference for other smart assistant applications.

CulinaX demonstrates a case of reasonably dividing the boundaries between deterministic logic and generative AI in mobile applications:

• Not all functions require large models, nor are all functions suitable for rule-based implementation
• Finding the optimal combination point between the two is essential to building applications that are both reliable and intelligent

For developers, this is a reference example for hybrid intelligent system design.