Ep 16: Knowledge Is Power — RAG Architecture & Embedding Fundamentals

graph TB
    subgraph "Without RAG"
        Q1["User: What's AgentUpdate's refund policy?"]
        Q1 --> LLM1["🤖 GPT-4o"]
        LLM1 --> A1["Sorry, I don't know ❌"]
    end
    subgraph "With RAG"
        Q2["Same question"]
        Q2 --> Search["🔍 Retrieve from KB
refund_policy.pdf → relevant chunks"]
        Search --> LLM2["🤖 GPT-4o + context"]
        LLM2 --> A2["Per our policy, full refund within 30 days... ✅"]
    end
    style Search fill:#22c55e,stroke:#16a34a,color:#fff

graph TB
    subgraph "Phase 1: Indexing (Offline)"
        Doc[📄 Documents] --> Chunk[✂️ Chunking]
        Chunk --> Embed[🧮 Embedding Model]
        Embed --> Store[💾 Vector Database]
    end
    subgraph "Phase 2: Retrieval (Online)"
        Query[❓ User Question] --> QEmbed[🧮 Question → Vector]
        QEmbed --> Sim[📐 Similarity Search]
        Sim --> TopK["📑 Top-K Chunks"]
        TopK --> LLM[🤖 LLM Generates Answer]
    end
    style Embed fill:#8b5cf6,stroke:#7c3aed,color:#fff
    style Sim fill:#f59e0b,stroke:#d97706,color:#fff