All About AI

A comprehensive collection of AI development resources, guides, and practical code examples

Created by: Beyhan MEYRALI

---

📖 About This Repository

Welcome to All About AI - your comprehensive resource hub for artificial intelligence development, covering everything from fine-tuning language models to setting up production-grade development environments.

This repository is designed for developers at all levels - whether you're taking your first steps into AI or building advanced production systems. Each guide is built from real-world experience and optimized for practical, hands-on learning.

---

🗂️ Repository Structure

🎯 fine-tuning/

The Great Fine-Tuning Revolution - A complete journey through AI model fine-tuning, from beginner to advanced.

What's Inside:

• 📚 00-first-time-beginner: Your first steps into AI fine-tuning
• ⚡ 01-unsloth: Fast and memory-efficient fine-tuning
• 🤗 02-huggingface-peft: Parameter-Efficient Fine-Tuning with HuggingFace
• 🦙 03-ollama: Local LLM deployment and usage
• 📦 04-quantization: Model compression techniques
• 💡 05-examples: Real-world use cases and implementations
• 🚀 06-advanced-techniques: Advanced optimization strategies
• 🔓 07-system-prompt-modification: Customizing model behavior
• 🏭 08-llamafactory: Production-grade fine-tuning framework

Hardware Focus: Optimized for AMD Ryzen 9 8945HS + Radeon 780M (GMKtec K11)

Learning Path: From 15-minute demos to production deployment

👉 Start Your Fine-Tuning Journey →

---

🖥️ perfect-setup/

Your Personal Remote Development Server - The game-changing setup that lets you run AI coding agents 24/7 in persistent tmux sessions, managed from any device.

🚀 The Revolutionary Workflow:

1. Assign Tasks to AI Agents - Start Claude Code, Copilot CLI, or Gemini CLI in tmux
2. Let Them Work - AI agents write code while you're away
3. Check Progress Anywhere - Review from phone, tablet, or any browser
4. Give Feedback & New Tasks - Test results, assign next tasks, repeat

Why This Changes Everything: Without AI agents, you need to be at your PC to write code. With this setup, AI agents write code for you while you guide, review, and test from anywhere. This is "vibe coding"—managing AI workers instead of writing code yourself.

What's Inside:

• 🤖 AI Coding Agents - Claude Code, GitHub Copilot CLI, Gemini CLI in persistent tmux sessions
• 📱 Remote Management - Check progress from phone, assign tasks from tablet
• 🌐 Access Anywhere - Full VS Code Server in browser, SSH from any device
• 🔐 Secure VPN - Tailscale for zero-config connections (no port forwarding!)
• 📂 Multi-Project - Multiple AI agents on different projects simultaneously
• 🐳 Complete Stack - WSL2, Docker, tmux, systemd, optional GPU acceleration

Perfect For: Any developer who wants AI agents to write code while they focus on architecture, review, and testing—accessible from anywhere.

Key Benefit: Your personal AI coding farm. Multiple agents working 24/7 on different projects. Zero ongoing costs. Manage from any device worldwide.

👉 Build Your AI-Powered Remote Development Server →

---

🤖 ai-agents/

AI Agents: From Zero to Hero - Learn to build production AI agents by understanding why you need frameworks, not just how to use them.

🔗 Featured on LinkedIn: Read the full announcement

🎯 The Philosophy:

Do you know exactly what happens when you send a message to ChatGPT? Most tutorials hide the complexity behind libraries. "Just import this framework and run."

I believe the best way to learn is to build it from scratch, locally, and watch the variables change in the debugger.

Don't use AI frameworks blindly. Learn why you need them first. 🛠️

🗺️ The Roadmap - From Raw Python to Production:

Most tutorials jump straight into complex frameworks, leaving you confused about what's actually happening. This repo takes the opposite approach.

We build from the ground up:

1. The Foundation: Raw HTTP calls and OOP Python classes ✅ Available now!
2. The Mechanics: Manual tool calling and recursion ✅ Available now!
3. The Realization: Understanding why manual state management gets messy
4. The Solution: Introducing Frameworks (LangChain, LangGraph, CrewAI) ✅ Available now!
5. The Knowledge: RAG Systems with vector databases ✅ Available now!
6. The Combination: Integrating RAG with Agent Frameworks ✅ Available now!
7. The Voice: Complete Voice Assistant ✅ Available now!
8. The Memory: Long-term context with Letta (MemGPT) 🚧 Coming soon

📚 What's Inside:

• 📚 00-llm-basics - Understanding stateless LLMs and API fundamentals
• 🔧 01-tool-calling - Function calling and recursive agent loops
• 🕸️ 02-agent-frameworks - LangChain, LangGraph & CrewAI (Modern LCEL)
• 📊 03-embeddings-rag - Vector databases and retrieval-augmented generation
• 🔗 04-integrated-agents - Combining RAG with CrewAI agent frameworks
• 🎙️ 05-voice-assistant - Complete voice assistant (Whisper + LangGraph + TTS)
• 🧠 04-memory-systems (Coming Soon) - Long-term memory with Letta (MemGPT)

🎓 How to Use This Repo:

1. Read the README in each folder for the theory
2. Run the Code with a debugger to see the practice
3. Study the Comments - Extensive explanations in every script

💻 Tech Stack: 100% Local with Ollama & Qwen. No API keys required.

🔍 Learning Philosophy:

• ✅ Build from scratch before using frameworks
• ✅ Debugger-friendly code with extensive comments
• ✅ curl examples for every HTTP endpoint
• ✅ Real-world examples from production systems
• ✅ Understand the "why" not just the "how"

👉 Start Building AI Agents →

---

📜 ai-history/

The Silicon God AI - Philosophical exploration of AI's evolution, impact, and future.

What's Inside:

• 🤖 The Silicon God AI - A deep dive into AI's transformation of society
• 🧠 Philosophy & Ethics - Understanding AI's role in human civilization
• 🌍 Future Perspectives - Where AI is taking us

Languages: Available in English and Turkish (Türkçe)

👉 Explore AI Philosophy →

---

⚡ llm-inference/

Run a 30B Model on an 8 GB Laptop GPU — measured at 53.8 tok/s. With reproducible commands and an explanation of why every flag exists, not just how to use it.

What's Inside:

• 📘 LESSONS_LEARNED.md — Feynman-style theory: VRAM hierarchy, MoE, quantization, KV cache, picking inference tools.
• 🛠️ HOW_TO_RUN.md — Every flag, every recipe, MoE tuning walkthrough, troubleshooting.
• 📊 BENCHMARKS.md — Measured tok/s table + TurboQuant long-context demo.
• ⚡ run_bench.sh — One-line llama-bench wrapper.
• 🌐 SERVING_AT_SCALE.md — Why serving an LLM is an OS problem. The KV-cache wall (one H100 holds ~29 conversations at 8K), PagedAttention as virtual memory, continuous batching as preemptive scheduling, prefill/decode disaggregation, and a claims audit of the viral version. With kv_cache_calc.py.
• 🎯 DIFFUSIONGEMMA.md — Jev vs DiffusionGemma vs Qwen (+ Ternary Bonsai 27B, Laya), measured. Bonsai is 5.9 GB, fits entirely on an 8 GB GPU, and is about as accurate as Jev on these tasks. Laya, a 421M open "Jev alternative", answers in 10–60 ms but is narrow. TypeSafe's Jev: flat 0.34 s, 94 % / 85 %, $0.014 per 1,000 decisions. DiffusionGemma as a Jev-style decision engine on an 8 GB laptop: 0 output tokens, 20 answers in one pass. Calibrated yes/no, choice and score answers from one bidirectional pass. Includes a llama.cpp backend for OpenJev, a 400-question benchmark against Qwen 3 30B-A3B (89 % vs 83 %, 0 vs 31 format errors), and a claims audit of the viral "~0.2 s, zero hallucinations" post.
• 🧮 LOCAL_CLUSTER.md — Buyer's guide to a local LLM cluster. Recommendations first, then the reasoning: whether you need a cluster at all, the three numbers that decide it, every option priced (used 3090s, Threadripper, Mac Studio, Mac mini clusters, DGX Spark vs RTX Spark, Strix Halo), the traps, and cluster_roofline.py to price your own shopping list before you pay.

Headline Results (RTX 5060 Laptop, 8 GB VRAM):

Hardware Focus: AMD Ryzen AI 9 365 + RTX 5060 Laptop GPU (8 GB, Blackwell sm_120). Built on llama.cpp from source + Madreag's TurboQuant fork.

👉 Start Running LLMs Locally →

---

🎯 Who Is This For?

🌱 Complete Beginners

• New to AI and machine learning
• Want to understand fine-tuning from scratch
• Looking for step-by-step guides with clear explanations

🌿 Intermediate Developers

• Familiar with Python and basic ML concepts
• Want to fine-tune models for specific tasks
• Setting up remote development environments

🌳 Advanced / AI Engineers

• Building production AI systems
• Optimizing model performance and deployment
• Implementing RAG pipelines and advanced techniques

---

🚀 Quick Start

For Fine-Tuning AI Models

cd fine-tuning/00-first-time-beginner
pip install -r requirements.txt
python test_setup.py

For Development Environment Setup

# On Windows PowerShell
wsl --install Ubuntu
# Then follow the guide in perfect-setup/

---

📚 What You'll Learn

AI Model Fine-Tuning

• ✅ How to customize large language models for your needs
• ✅ Memory-efficient training techniques (LoRA, QLoRA)
• ✅ Quantization and model compression
• ✅ Deployment strategies (Ollama, vLLM)
• ✅ Building practical AI applications

Development Environment

• ✅ Setting up WSL2 for AI development
• ✅ Remote access and persistent sessions (tmux)
• ✅ Running multiple AI agents simultaneously
• ✅ GPU acceleration for AI workloads
• ✅ Secure networking with Tailscale VPN

---

💡 Philosophy

This repository is built on three core principles:

1. Practical First: Every guide is tested on real hardware with real use cases
2. Beginner-Friendly: Complex concepts explained with analogies and clear examples
3. Production-Ready: Not just demos - techniques you can use in real projects

---

🛠️ Hardware & Software

Fine-Tuning Optimization

• Primary: AMD Ryzen 9 8945HS + Radeon 780M (GMKtec K11)
• GPU: AMD ROCm for GPU acceleration
• Memory: Optimized for 32GB RAM + 8GB shared GPU memory

Development Environment

• Primary: AMD Ryzen AI 9 365 + RTX 5060 Laptop GPU
• OS: Windows 11 with WSL2 Ubuntu 24.04 LTS
• GPU: NVIDIA CUDA for AI inference

Software Stack

• Python 3.11+
• PyTorch 2.0+
• HuggingFace Transformers & PEFT
• Unsloth, LlamaFactory
• Docker, vLLM, Ollama, Qdrant

---

📖 Additional Resources

Documentation in Turkish 🇹🇷

• fine-tuning/README-TR.md - Turkish version of fine-tuning guide
• ai-history/The_Silicon_God_AI-TR.md - Silicon God AI in Turkish

Deep Dives

• The Silicon God AI - Philosophy and future of AI
• CLAUDE.md - Working with Claude AI

---

🤝 Contributing

Found a bug? Have a suggestion? Want to add your own guides?

1. Fork the repository
2. Create your feature branch
3. Submit a pull request

All contributions are welcome!

---

📫 Connect

Beyhan MEYRALI

• 💼 LinkedIn
• 🐙 GitHub

---

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

---

⭐ Star This Repository

If you find this repository helpful, please consider giving it a star! It helps others discover these resources.

---

Happy Learning! 🚀

"The best way to predict the future is to build it."