# Technical Document Extraction: RAG / Iterative RAG Prompt Structure
## 1. Overview
This image is a technical diagram illustrating the architectural composition of a prompt used for Retrieval-Augmented Generation (RAG) or Iterative RAG. It demonstrates how multiple in-context examples and test data are sequenced to generate a final answer.
## 2. Component Isolation
### Region A: Header / Global Brackets
* **Top Bracket:** Spans the first two sets of examples. Label: `In-context examples (m)`.
* **Bottom Bracket:** Spans the entire sequence from the first example to the test query. Label: `Prompt`.
### Region B: Main Sequence (Input Data)
The input is composed of repeating blocks and a final test block, color-coded by function:
* **Red Blocks:** Represent source material. Label: `Documents (k)`.
* **Blue Blocks:** Represent queries.
* **Green Blocks:** Represent answers.
#### Sequence Breakdown:
1. **Example Set 1:**
* **Red Block:** `Example Documents 1` (Sub-labeled `Documents (k)`)
* **Blue Block:** `Example Query 1`
* **Green Block:** `Example Answer 1`
2. **Ellipsis (`...`):** Indicates a continuation of the pattern for $n$ examples.
3. **Example Set n:**
* **Red Block:** `Example Documents n` (Sub-labeled `Documents (k)`)
* **Blue Block:** `Example Query n`
* **Green Block:** `Example Answer n`
4. **Test Set:**
* **Red Block:** `Test Documents` (Sub-labeled `Documents (k)`)
* **Blue Block:** `Test Query`
### Region C: Processing and Output
* **Icon:** A stylized robot head representing an AI model/LLM.
* **Transition Arrow:** A grey arrow pointing from the `Prompt` sequence to the `Final Answer`.
* **Process Label:** `Generate with RAG / Iterative RAG`.
* **Output Block (Green):** `Final Answer`.
## 3. Data Structure and Logic
The diagram defines a specific mathematical and logical relationship for the prompt construction:
| Component | Variable | Description |
| :--- | :--- | :--- |
| **In-context examples** | $m$ | The total number of few-shot examples provided in the prompt. |
| **Documents** | $k$ | The number of retrieved documents provided for each specific query (both example and test). |
| **Prompt** | N/A | The concatenation of $m$ examples (Document + Query + Answer) followed by the Test Document and Test Query. |
## 4. Flow Description
1. **Contextualization:** The system retrieves $k$ documents for $m$ different examples. Each example consists of the documents, a query, and the correct answer.
2. **Test Input:** The system retrieves $k$ documents for the current `Test Query`.
3. **Prompt Assembly:** All $m$ examples and the current test data are bundled into a single `Prompt`.
4. **Generation:** The AI model processes the `Prompt` using a `RAG` or `Iterative RAG` methodology.
5. **Termination:** The process results in a single `Final Answer`.
## 5. Text Transcription (Precise)
* **Top Labels:** "In-context examples (m)", "Documents (k)" (repeated 3 times).
* **Internal Block Text (Vertical):**
* "Example Documents 1"
* "Example Query 1"
* "Example Answer 1"
* "Example Documents n"
* "Example Query n"
* "Example Answer n"
* "Test Documents"
* "Test Query"
* "Final Answer"
* **Action Text:** "Generate with RAG / Iterative RAG"
* **Bottom Label:** "Prompt"
