Note: This project is not an agentic AI system. It does not rely on autonomous agents, planners, or multi-step self-directed behaviors. Instead, it uses a simple, “agentless” architecture where the backend interacts directly with LLMs without delegating tasks to independent agents.
This web application uses AI to analyze recipe pages and extract structured ingredients.
This is a small learning project created to explore AI-driven solutions and gain a deeper understanding of how LLMs and modern AI concepts work, helping me prepare for the new trends in software development.
👉 AI Agentless Recipe & Shoplist on GitHub
🧠 What It Does
An intelligent Python 3.11+ application that analyzes recipe websites, extracts ingredients using AI, and searches multiple grocery stores for matching products. This project demonstrates AI-powered recipe analysis with real-world grocery store integration.
This “agentless” version focuses on generating a complete cooking flow using only Python + LLM calls — no agents, no complex orchestrators. You give it ingredients, and it produces everything you need:
- 🤖 AI-Powered Recipe Extraction: Uses OpenAI, Azure OpenAI, Ollama, or GitHub Models to intelligently parse recipe websites.
- 🛒 Multi-Store Product Search: Searches Coles, Woolworths, ALDI, and IGA with real product URLs.
- 📝 Full Shopping List Generation: Generates a complete shopping list based on an online recipe.
- 🏆 Best Product Selection: Selects the best matching product from multiple stores based on quality and price.
Example recipe processing:
Using Gazpacho from RecipeTin Eats — demonstrating AI-powered ingredient extraction from live recipe websites.
This is an example of the structured output produced by the script after analyzing the recipe and searching grocery stores:
💡 Why I Built It
I had already been experimenting with recipe crawlers and automated shopping lists, but I wanted something simpler and more focused: a tool that takes only the ingredients you have and generates the rest using AI reasoning. At the same time, I saw it as an opportunity to learn AI-based solutions, understand how LLMs work, and explore other AI concepts in a hands-on way.
- I wanted a lightweight, deterministic approach — no agents running loops.
- It’s a fun way to test prompt engineering + structured output.
- It’s a tool I can reuse every week to plan meals quickly.
🔍 How the AI Recipe Shopping List Project Works
This project is an intelligent web application that uses AI to analyze recipes, extract ingredients, and find matching products across multiple grocery stores. It combines a Python backend, AI providers, store integrations, and a modern frontend to create a seamless shopping experience.
📘 System Overview
- User Input: The user enters a recipe URL into the web interface.
- AI Extraction: The backend extracts ingredients, quantities, and instructions using an AI provider (OpenAI, Azure, Ollama, or GitHub Models).
- Store Search: The system queries Coles, Woolworths, ALDI, and IGA for matching grocery items.
- Product Matching: AI selects the best product based on price, size, and relevance.
- Results Display: The frontend shows matched products, store links, and a structured shopping list.
- Demo Mode: Users can try the app with a preloaded sample recipe and real product data.
🏗️ Architecture Diagram
🔄 Data Flow
- Input: The user provides a recipe URL.
- AI Extraction: Backend sends the URL to an AI provider to obtain structured recipe data.
- Store Search: Each ingredient triggers a query to store APIs or scrapers.
- Product Matching: AI logic selects the best item for each ingredient.
- Frontend Display: The results are shown with product images, prices, and store links.
🧩 Key Components
- FastAPI Backend: Manages recipe extraction, store search, and all business logic.
- AI Providers: Used for extraction and product matching.
- Grocery Store Integrations: Fetch real product data from Coles, Woolworths, ALDI, and IGA.
- Frontend (HTML/JS): Provides the user interface for inputs and results.
- Storage System: Caches and stores data for reliability and speed.
➡️ Example Workflow
- User enters a recipe URL and clicks Process Recipe.
- The backend extracts recipe details using AI.
- The system searches multiple stores for matching ingredients.
- AI picks the best matches.
- The frontend displays a structured shopping list with links to products.
For more details, see the README and documentation included in the project repository.
📚 Conclusion: What I Learned
This project was a practical way to deepen my understanding of modern AI development without relying on agentic architectures. By building everything using a simple, agentless workflow, I learned how to combine LLMs, backend services, and real-world data sources in a clean and controlled way.
- AI & LLM Fundamentals: Gained hands-on experience with how LLMs interpret unstructured text (like recipe pages) and transform it into reliable structured data.
- Token Awareness: Learned how token limits, context windows, and prompt size influence output quality — essential when working with different models and optimizing costs.
- Using Free GitHub Models: Explored how to run capable LLMs at zero cost through GitHub Models, allowing rapid experimentation, iteration, and testing without worrying about usage fees.
- Prompt Design: Discovered how small adjustments in prompt wording significantly improve extraction accuracy and product-matching quality.
- Backend & API Integration: Built a FastAPI backend that coordinates recipe extraction, store product searches, and AI-driven matching logic.
- Working with Real-World Data: Learned the challenges of handling messy grocery store data, inconsistent website structures, and normalizing product information.
Overall, this project strengthened my understanding of LLM behavior, prompt engineering, and end-to-end AI integration — and it laid the foundation for future explorations into agentic AI workflows.
🚀 What's Next
The next phase of this project will evolve into a full agentic AI workflow, moving beyond the current agentless architecture. The goal is to explore how autonomous agents, orchestration, and multi-step reasoning can coordinate a more advanced recipe-to-shopping pipeline.
- Multi-Agent Architecture: Introduce specialized agents for tasks like recipe extraction, ingredient normalization, store product search, and shopping-list optimization.
- Orchestrated AI Workflow: Implement a controller/orchestrator agent capable of delegating work to sub-agents, validating their outputs, and managing the full end-to-end reasoning process.
- Model Context Protocol (MCP): Move website scraping into dedicated MCP tools so agents can call them in real time, enabling clean, secure, and maintainable integrations.
- AI Collaboration: Allow multiple agents to collaborate — for example, an extraction agent talking to a store-matching agent — improving accuracy and autonomy.
- Richer Decision-Making: Use agentic reasoning to evaluate factors like product quality, quantity equivalence, pricing, and availability in a more sophisticated way.
This next version will serve as a hands-on experiment in building real agentic systems, helping me deepen my understanding of orchestration, tool calling, autonomous workflows, and the future of AI-driven application design.
🤝 Want to Contribute?
If this sounds interesting, feel free to:
- Fork the repo and experiment.
- Open issues with ideas or improvements.
- Share how you’d use this — I’m always curious to see new workflows!
){kind=link}