ChefMind AI

    Intelligent Kitchen Recommendation System

    A full-stack AI application I built from scratch, combining vector embeddings, semantic search, and hybrid ranking algorithms to solve the daily "what's for dinner?" problem.

    Python
    Node.js
    OpenAI Embeddings
    Pinecone
    AWS S3
    JWT Auth
    React Native
    Netlify
    01

    The Insight

    Have you ever noticed something fascinating? Same kitchen. Same ingredients. Yet one person commits an atrocity against the human tongue while another pulls together a gourmet meal that brings the whole family to the table.

    The difference isn't the ingredients, it's knowing what to do with them.

    "The difference is knowing what to do with them."

    02

    The Problem

    Every evening, millions of people open their fridge and ask: "What's for dinner?" They have tomatoes, chicken, half an onion, and mysterious leftovers. The meal is in there somewhere, but the mental load of figuring it out turns cooking from joy into a chore.

    The Solution

    Upload your kitchen inventory. Get AI-powered recipe recommendations ranked by what you actually have, what you like, and what fits your dietary needs.

    03

    About the Developer

    I designed and built this entire system, from the embedding pipeline to the mobile app. This project showcases my ability to work across the full AI application stack.

    What I Built:

    • Vector embedding pipeline with OpenAI's text-embedding-3-small
    • Production vector database migration (ChromaDB → Pinecone)
    • RESTful API with JWT authentication and secure password hashing
    • Hybrid scoring algorithm balancing relevance + availability
    • Cross-platform delivery: CLI, React Native mobile, React PWA
    • Serverless deployment on Netlify Functions

    App Demo

    Experience the application directly below. Try interacting with the interface to see how ChefMind AI helps manage your kitchen and suggests recipes.

    04

    How It Works

    Ingredients go in, ranked recipe matches come out. Follow the engine from your kitchen inventory to the final ranking.

    Kitchen Inventory
    Semantic Embeddings
    Vector Search
    Smart Ranking
    1. 01

      Kitchen Inventory

      You upload what is actually in your kitchen. Chicken, tomatoes, rice, garlic, and whatever else is on hand becomes the starting point.

      Ingredients in

    2. 02

      Semantic Embeddings

      Each ingredient is converted into a vector with OpenAI text-embedding-3-small. Unlike keyword matching, the AI understands that chicken breast relates to poultry dishes and that tomatoes work in both Italian and Mexican cuisine.

      Meaning, not keywords

    3. 03

      Vector Search

      Pinecone searches the recipe space for the closest matches to your ingredient vectors, surfacing dishes you can realistically make.

      Closest recipes surface

    4. 04

      Smart Ranking

      A hybrid scoring algorithm ranks the results, weighting semantic relevance at 70 percent and availability at 30 percent, so the top suggestions are both a good fit and actually cookable right now.

      Ranked by fit and availability

    Recipe Match Engine

    From ingredients to ranked matches

    01
    Kitchen Inventory
    02
    Semantic Embeddings
    03
    Vector Search
    04
    Smart Ranking

    Input

    chicken, tomatoes, rice, garlic, soy sauce, eggs, spinach, coconut milk...

    Output

    Chicken Stir Fry, 94% Match, Ready to cook!

    Coconut Curry Rice, 91% Match, Ready to cook!

    Egg Fried Rice, 85% Match, Missing: green onions

    Hybrid Scoring Algorithm

    70%
    Semantic Relevance
    How well do your ingredients match?
    30%
    Availability Score
    Can you actually make this right now?
    05

    Full Feature Set

    AI Recipe Matching

    Semantic search finds recipes that fit your ingredients

    Inventory Management

    Track quantities, auto-deduct when you cook

    Shopping List Generation

    Auto-generate lists based on low stock & recipes

    User Authentication

    Secure accounts with JWT & encrypted passwords

    Dietary Preferences

    Filter by allergies, restrictions, favorite cuisines

    Recipe Images

    Beautiful food photography via AWS S3

    Serverless API

    Production-ready deployment on Netlify

    Cross-Platform

    CLI, Mobile, Web, and API interfaces

    06

    Technical Implementation

    ComponentTechnology
    Vector DatabasePinecone (production-scale)
    EmbeddingsOpenAI text-embedding-3-small
    Backend APINode.js + Express
    AuthenticationJWT + bcrypt
    Cloud StorageAWS S3
    DeploymentNetlify Functions (serverless)
    MobileReact Native + Expo
    WebReact PWA + Material-UI
    07

    API Architecture

    API Endpoints
    POST /api/auth/signup Create account
    POST /api/auth/login JWT authentication
    GET /api/inventory View kitchen stock
    POST /api/inventory/restock Add items after shopping
    POST /api/recommendations Get AI-matched recipes
    POST /api/recipes/accept Cook & auto-deduct ingredients
    POST /api/shopping-list/generate Smart shopping list
    08

    Performance

    0+
    Recipes
    0
    Ingredients
    <0s
    Query response
    0%+
    Accuracy
    0
    Platforms
    140+

    Recipes

    14

    Food Groups

    233

    Ingredients

    <2s

    Query Response

    92%+

    Accuracy

    4

    Platforms

    09

    Skills Demonstrated

    Backend & API

    • • RESTful API design (Express.js)
    • • JWT authentication & middleware
    • • Secure password hashing (bcrypt)

    AI/ML Engineering

    • • Vector embeddings & semantic search
    • • Pinecone vector database
    • • OpenAI API integration
    • • Hybrid ranking algorithms

    Cloud Infrastructure

    • • AWS S3 for asset storage
    • • Serverless deployment (Netlify)
    • • Environment-based configuration

    Full-Stack Delivery

    • • React Native mobile app
    • • Progressive Web App
    • • Cross-platform architecture
    10

    What I Learned

    Vector databases for semantic similarity search
    Embedding pipelines with OpenAI models
    Hybrid scoring algorithms balancing relevance + practicality
    Full-stack architecture across CLI, mobile, and web
    Production deployment with serverless functions
    Secure authentication patterns (JWT + bcrypt)
    Cloud asset management with AWS S3
    4DAnalytics

    © 2025 4D Analytics | Isaacson Shoko

    Independent AI Systems Consultant | Johannesburg, South Africa