yeyu2/Youtube_demos

What it solves

This repository is a practical reference for developers who want to build real-world AI applications without guessing. Instead of reading scattered documentation, you get working code examples that accompany each YouTube video, covering topics from multi-agent frameworks to realtime voice and multimodal assistants. It solves the problem of 'where do I start' by providing ready-made, copy-pasteable code that matches exactly what the video explains.

How it works

The repo is organized into folders and files, each with a one-line description and a YouTube link. Each folder or script demonstrates a concrete technique or tool: for example, agent-context-isolation/ shows how to stop an AI coding agent from reviewing its own code, adk-rag/ shows building a voice RAG agent with ADK and File Search, and Llama32_FunctionCall/ shows function calling for lightweight Llama models. The code appears to be the exact code used in the corresponding video, so viewers can follow along or reuse it directly. Many entries focus on Google's ADK and Gemini models, while others use AutoGen, CrewAI, or open-source models like Llama, Gemma, and Phi. The links go to YouTube videos that explain the concepts in detail.

Who it’s for

This is for AI developers, hobbyists, and engineers who prefer learning by example. It suits anyone who wants to quickly prototype with current AI agent frameworks, realtime multimodal APIs, or local model deployments, and who appreciates having a concrete codebase to run, modify, and understand rather than just reading theoretical docs.

Highlights

• A wide range of examples: from realtime voice chat apps to RAG pipelines, agent orchestration, and local multimodal systems
• Every entry is paired with a YouTube tutorial, making it a video-plus-code learning resource
• Covers popular frameworks and APIs: Gemini, ADK, AutoGen, CrewAI, LiveKit, Ollama, and more
• Includes both cloud-based and fully local setups, as well as lightweight model options like Llama 3.2 and Phi-4