🟠 Stage 2: Intermediate

Goal: Learn to use and customize AI models effectively

Welcome to the Intermediate stage! Now that you understand how AI works under the hood, it's time to learn how to work with models effectively in real-world applications.

📚 Concepts in This Stage

graph TB
    A[Foundation Complete] --> B[Fine-tuning]
    A --> C[Few-shot Prompting]
    B --> D[Context Engineering]
    C --> D
    D --> E[Chain of Thought]
    E --> F[Vector Database]
    F --> G[RAG]
    
    style B fill:#FF9800
    style C fill:#FF9800
    style D fill:#FF9800
    style E fill:#FFC107
    style F fill:#FFC107
    style G fill:#4CAF50

7. Fine-tuning 🎯

Time: 5-6 hours | Difficulty: ⭐⭐⭐⭐☆

Adapt pre-trained models to your specific needs. Learn when to fine-tune vs when to use prompting.

Key Takeaways:

• When and why to fine-tune
• Preparing training data
• Fine-tuning with HuggingFace
• LoRA and efficient fine-tuning
• Evaluating fine-tuned models

Code Examples:

• Fine-tune BERT for classification
• Fine-tune GPT for specific style
• Use LoRA for efficient fine-tuning
• Evaluate model performance

8. Few-shot Prompting 📝

Time: 2-3 hours | Difficulty: ⭐⭐☆☆☆

Teach models through examples in the prompt. Master the art of in-context learning.

Key Takeaways:

• Zero-shot vs few-shot vs many-shot
• Selecting good examples
• Example ordering matters
• Dynamic example selection
• When prompting beats fine-tuning

Code Examples:

• Build few-shot classifiers
• Dynamic example selection
• Example retrieval systems
• Prompt optimization

9. Context Engineering 🔧

Time: 3-4 hours | Difficulty: ⭐⭐⭐☆☆

Master the art of crafting effective prompts. Learn advanced techniques for getting the best results from LLMs.

Key Takeaways:

• Prompt structure and formatting
• System prompts and roles
• Temperature and sampling
• Prompt templates and patterns
• A/B testing prompts

Code Examples:

• Build prompt templates
• Implement prompt optimization
• Create prompt libraries
• Measure prompt effectiveness

10. Chain of Thought 🧩

Time: 3-4 hours | Difficulty: ⭐⭐⭐☆☆

Make models reason step-by-step. Dramatically improve performance on complex tasks.

Key Takeaways:

• What is chain-of-thought
• Zero-shot CoT ("Let's think step by step")
• Few-shot CoT with examples
• Self-consistency
• Tree of thoughts

Code Examples:

• Implement CoT prompting
• Build reasoning chains
• Self-consistency voting
• Complex problem solving

11. Vector Database 🗄️

Time: 4-5 hours | Difficulty: ⭐⭐⭐⭐☆

Store and retrieve embeddings efficiently. Essential for semantic search and RAG systems.

Key Takeaways:

• What are vector databases
• Similarity search algorithms (ANN)
• Popular vector DBs (Pinecone, Weaviate, ChromaDB)
• Indexing strategies
• Performance optimization

Code Examples:

• Use ChromaDB locally
• Build semantic search
• Implement hybrid search
• Optimize retrieval performance

12. Retrieval Augmented Generation 🔍

Time: 5-6 hours | Difficulty: ⭐⭐⭐⭐⭐

Combine retrieval with generation. Build AI systems that can access external knowledge.

Key Takeaways:

• What is RAG and why it matters
• RAG pipeline components
• Chunking strategies
• Retrieval optimization
• Re-ranking and filtering

Code Examples:

• Build complete RAG system
• Implement with LangChain
• Optimize chunk size
• Add metadata filtering
• Evaluate RAG performance

🎯 Learning Path

Recommended Order

1. Fine-tuning - Understand model customization
2. Few-shot Prompting - Learn in-context learning
3. Context Engineering - Master prompt crafting
4. Chain of Thought - Enable reasoning
5. Vector Database - Store embeddings
6. RAG - Combine retrieval + generation

Time Commitment

• Total: 25-30 hours
• Per Day: 2-3 hours recommended
• Duration: 2-3 weeks at steady pace

Prerequisites

• Completed Foundation Stage
• Comfortable with Python
• Access to LLM APIs (OpenAI, Anthropic) or local models
• Basic understanding of databases

💻 Hands-On Projects

After completing this stage, build one of these projects:

🎮 Project 1: Custom Q&A System

Build a RAG-based question-answering system for your documents.

• Time: 6-8 hours
• Skills: RAG, Vector DB, Context Engineering
• Output: Working Q&A system with web interface

🎮 Project 2: Code Assistant

Create an AI coding assistant with context awareness.

• Time: 5-7 hours
• Skills: Fine-tuning, Few-shot, CoT
• Output: VS Code extension or CLI tool

🎮 Project 3: Research Assistant

Build an AI that can search, summarize, and synthesize information.

• Time: 8-10 hours
• Skills: RAG, Vector DB, Chain of Thought
• Output: Research tool with citations

📊 Progress Tracking

• Fine-tuning - Completed exercises
• Few-shot Prompting - Completed exercises
• Context Engineering - Completed exercises
• Chain of Thought - Completed exercises
• Vector Database - Completed exercises
• RAG - Completed exercises
• Built Project 1, 2, or 3

🔗 Quick Navigation

Previous: Foundation Stage Next: Advanced Stage Start: Begin with Fine-tuning