ChefRserve

A recipe finder that works from what you already have. Enter the ingredients in your pantry, and ChefReserve searches a database of real recipes, ranks them by how many ingredients you already own.

Comment

Inspiration

ChefReserve was inspired by a common problem faced by students and busy individuals: deciding what to cook with limited ingredients, time, and decision bandwidth.

Instead of browsing endless recipes, users often want a simple answer to:

“What can I actually cook right now with what I already have?”

I wanted to solve this by combining structured recipe data with AI-powered ranking and personalization to make cooking faster, smarter, and more accessible.

What it does

ChefReserve is an AI-powered recipe intelligence platform that transforms pantry ingredients into ranked, personalized meal suggestions.

It:

  • Matches user-provided ingredients with a large recipe database
  • Ranks recipes based on ingredient overlap and availability
  • Uses AI to refine and rerank results based on preferences (cuisine, difficulty, cook time, spice level)
  • Highlights missing vs available ingredients for each recipe
  • Links users to original recipe sources for full instructions

The system turns fragmented recipe search into a guided, personalized cooking experience.

How I built it

ChefReserve is a full-stack system combining web development, data engineering, and LLM-based ranking:

  1. Frontend: React + Tailwind CSS for a responsive, interactive UI
  2. Backend: Flask API handling search, filtering, and ranking logic
  3. Database: Firebase Firestore storing structured recipe data with indexed ingredients
  4. AI Layer: Groq (Llama 3.1 8B) used for:
  5. Semantic reranking of recipe results
  6. Enriching scraped recipes with metadata (cuisine, difficulty, spice level)
  7. Data Pipeline: Custom Python web crawler using BeautifulSoup to extract recipe data from structured 8. JSON-LD (schema.org/Recipe) across multiple cooking sites
  8. Search Engine: Ingredient-token matching system that computes recipe relevance scores before AI reranking

The system is designed as a pipeline: User input → ingredient matching → candidate retrieval → AI reranking → filtered output

Challenges I ran into

  1. Designing a robust ingredient matching system that handles noisy or incomplete user input
  2. Preventing AI hallucination while still leveraging LLMs for ranking and enrichment
  3. Structuring unstandardized recipe data from multiple websites into a consistent schema
  4. Balancing deterministic ranking (ingredient overlap) with probabilistic AI reranking
  5. Keeping the UI simple while supporting a multi-stage recommendation pipeline

Accomplishments that I am proud of

  1. Built a fully functional end-to-end AI-powered recipe intelligence system
  2. Integrated LLM-based reranking into a real-world recommendation pipeline
  3. Designed a scalable recipe ingestion system using web scraping + structured extraction
  4. Created a system that meaningfully reduces decision fatigue in cooking
  5. Delivered a production-style full-stack application under hackathon constraints

What I learned

  1. How to integrate LLMs into structured recommendation pipelines (not just chatbots)
  2. The importance of deterministic + probabilistic hybrid systems in AI applications
  3. Real-world challenges of data cleaning and schema normalization
  4. Trade-offs between AI creativity and system reliability
  5. How to design user-centric systems under time constraints

What's next for ChefReserve

  1. Add feedback loops to improve ranking based on user behavior
  2. Introduce user accounts and personalized recommendation history
  3. Build grocery list generation from selected recipes
  4. Expand recipe dataset with more cuisines and dietary restrictions
  5. Explore lightweight meal planning and weekly scheduling features

Built With

Share this project:

Updates