Add 08 content

Author: girliemac

08-multimodal-multimodel/README.md

@@ -1 +1,107 @@
-Doodle summary
+# 🧠 Multimodal and Multimodel in AI
+
+Let's learn about two concepts you'll hear a lot in modern AI — **multimodal** and **multi-model**.
+They sound almost the same, but they work in very different ways.
+
+![Multimodal and Multi-model AI](../images/multimodal-multimodel.png)
+
+## 🧿 What Is Multimodality?
+
+**Multimodality** means an AI model can understand and generate *multiple types of data* and combine them to reason about the world.
+
+A **multimodal model** is a *single AI model* that works across:
+
+- 📝 Text  
+- 🖼️ Images  
+- 🔊 Audio  
+- 🎥 Video  
+- 💻 Code  
+etc.
+
+Think of it as **one brain with multiple senses** — all working together.  
+The model can read an image, interpret the text inside it, listen to accompanying audio, and respond in natural language, all in one flow.
+
+### When multimodal model approaches shine
+- Simple, end-to-end tasks  
+- Scenarios where reasoning across multiple data types matters  
+- Apps where you want **one model** and minimal engineering overhead  
+- Fast prototyping or lightweight workflows
+
+
+## 👯 What Is the Multi-model Approach?
+
+A **multimodel** system uses **multiple specialized AI models**, each designed for a specific task.
+
+Examples:
+- 👁️ A vision model for image understanding  
+- 🌐 A translation model for languages  
+- 💬 A large language model for reasoning  
+- 🧩 A classifier or embedding model for structured tasks  
+
+Your application becomes the **orchestrator**, passing outputs from one model to another as a workflow.
+
+This is like having a **team of experts**, each doing what they're best at.
+
+### When multi-model systems shine
+- High-accuracy, domain-specific requirements  
+- Workflows that need fine control at each step  
+- Combining best-in-class models for each modality  
+- Large-scale pipelines where cost efficiency matters  
+
+## ⚖️ Which Approach Should You Choose?
+
+**Multimodal**  
+- ✔ Fewer moving parts  
+- ✔ Easy to build with  
+- ✔ Great for general use  
+- ❌ Can be more expensive per inference  
+- ❌ Not always the best for specialized tasks  
+
+**Multimodel**  
+- ✔ Higher accuracy through specialization  
+- ✔ More cost-efficient at scale  
+- ✔ Fine-grained control  
+- ❌ Requires more engineering  
+- ❌ More points of failure  
+
+## 🧜‍♀️ Hybrid Approaches (Often the Sweet Spot)
+
+In many real applications, you'll mix both:
+
+- Use **specialized models** for tasks like OCR or transcription  
+- Use a **multimodal model** or LLM on top to reason and produce a final answer  
+
+This gives you a balance of accuracy, cost efficiency, and flexibility.
+
+
+## 🧪 Example in the video: Multi-model Approach
+
+**Scenario:**
+You're traveling in Japan, sitting at a restaurant for lunch.
+You take a photo of the menu and ask your app:
+
+> "Can you suggest gluten-free meals from this menu?"
+
+![Diner in Japan](../images/japan-diner.png)
+
+**How the app handles it:**
+
+The app orchestrates three specialized models, each doing what it's best at:
+
+1. **OCR Model** extracts the text from the menu image — including Japanese characters, prices, dish names, and descriptions.
+1. **Translation Model** translates the extracted text into English (or the user's preferred language) with high linguistic accuracy.
+1. **LLM for Reasoning** analyzes the translated menu, identifies ingredients, checks for gluten-containing items, and returns a clear recommendation of safe dishes.
+
+---
+
+## 🚀 Try the Example App
+
+**Now [Try the Example App by Yourself](sample/README.md)!**
+
+## 📺 Watch on YouTube - Will be available soon!
+
+Watch the video, **Multimodal and Multi-model AI** on YouTube:
+
+[![YouTube: Multimodal and Multi-model AI](https://img.youtube.com/vi/0000/0.jpg)](https://www.youtube.com/watch?v=0000)
+
+[Subscribe us!](https://www.youtube.com/channel/UCV_6HOhwxYLXAGd-JOqKPoQ?sub_confirmation=1)

08-multimodal-multimodel/sample/README.md

@@ -97,6 +97,8 @@ Access the app in your browser at:
 
 [http://localhost:3001](http://localhost:3001)
 
+Download [this Japanese diner menu](../../images/contoso-lunch.png) and try with the app!
+
 ## 🧠 How It Works
 
 The app employs a tiered router-based architecture to handle multimodal and multimodel tasks: