---
title: Multimodal Recipe RAG
emoji: 🍳
colorFrom: red
colorTo: pink
sdk: gradio
sdk_version: 4.44.1
app_file: app.py
pinned: false
license: apache-2.0
tags:
  - multimodal
  - rag
  - recipe-search
  - nvidia-nemotron
  - computer-vision
  - food
---

# 🍳 Multimodal Recipe RAG System

Search **500 recipes** using text queries across images and documents with three powerful search modes:

- 🖼️ **Image**: Visual similarity search (find similar-looking dishes)
- 📝 **Text**: Semantic text search (find by ingredients, instructions, reviews)
- 🎨 **Image+Text**: Combined multimodal search (best of both worlds)

## 🔧 Technology

**Models:**
- Embedding: [nvidia/llama-nemotron-embed-vl-1b-v2](https://huggingface.co/nvidia/llama-nemotron-embed-vl-1b-v2)
- Reranking: [nvidia/llama-nemotron-rerank-vl-1b-v2](https://huggingface.co/nvidia/llama-nemotron-rerank-vl-1b-v2)

**Dataset:** [TurkishCodeMan/recipe-synthetic-images-10k](https://huggingface.co/datasets/TurkishCodeMan/recipe-synthetic-images-10k)

**Embeddings:**
- Image: 2048-dim visual embeddings
- Text: 2048-dim semantic embeddings  
- Image+Text: 2048-dim multimodal embeddings

## 💡 Usage

1. Enter your search query (e.g., "chocolate cake", "healthy breakfast")
2. Choose search modality (Image/Text/Image+Text)
3. Optionally enable reranking for better results
4. Click Search!

## 🎯 Example Queries

**Image Mode:**
- "recipes with steak"
- "chocolate desserts"
- "colorful salads"

**Text Mode:**
- "healthy breakfast ideas"
- "vegetarian meals with pasta"
- "quick dinner under 30 minutes"

**Image+Text Mode (Recommended):**
- "creamy pasta dishes"
- "spicy chicken recipes"
- "fresh summer salads"

## ⚡ Performance

- Average search time: ~1-2 seconds
- With reranking: ~2-3 seconds
- GPU recommended for optimal performance

## 🏗️ Architecture
Query → Embedding Model → Similarity Search → [Optional Reranking] → Top Results



## 📊 Dataset

500 recipes from TurkishCodeMan/recipe-synthetic-images-10k:
- Recipe images (PNG)
- Full recipe text (markdown)
- Ingredients, steps, nutrition info
- User reviews

## 🙏 Credits

Built with:
- NVIDIA Nemotron models
- Hugging Face Transformers
- Gradio for UI

# Yemek temalı
colorFrom: red
colorTo: pink

# Alternatifler
colorFrom: yellow
colorTo: red

colorFrom: purple
colorTo: pink

colorFrom: blue
colorTo: indigo

colorFrom: green
colorTo: blue