Split long documents into semantically complete text chunks, 200-500 tokens each, with 10-20% overlap between adjacent chunks
↓
2
🔢 Vectorization (Embedding)
Convert each text chunk into a high-dimensional vector (e.g., 1536 dimensions) via an Embedding model. Semantically similar text is closer in vector space.
↓
3
🔍 Query Vectorization + Similarity Search
User question: "What are the applications of Python in AI?" → Vectorize → Find the nearest Top-K chunks in vector space
↓
4
🧩 Prompt Assembly
Inject retrieved text chunks into the System Prompt to build a complete augmented context
[System] You are a professional assistant. Please answer questions based on the following reference materials.
[Context] Chunk 3: "Python is widely used in artificial intelligence, including machine learning (TensorFlow, PyTorch), natural language processing (Transformers), computer vision (OpenCV)..." Chunk 7: "Python's scikit-learn library provides rich traditional ML algorithms..."
[User] What are the applications of Python in AI?
↓
5
🤖 LLM Generates Answer
The large model generates an answer based on augmented context, ensuring grounded responses and reducing hallucinations