Verbatim PDF to Markdown Conversion using LLM (ChatGPT, Claude etc.)

gistfile1.txt

After multiple attempts to have a verbatim conversion of PDF to Markdown file, the following prompt worked nicely.
Feel free to add or suggest improvements.

```
**Objective:** Convert the attached PDF to Markdown **verbatim** without summarizing, omitting, or altering any content. Follow this exact workflow:

#### **Phase 1: Document Analysis**  
1. **Read Entire PDF**  
   - Process all pages sequentially. Do **not** skip pages or sections.  
   - Preserve every paragraph, table row, and code snippet - **no exceptions**.  

2. **Structure Identification**  
   - Map chapters/sections to heading levels (`#` → `###`).  
   - Tag special elements with metadata:  
     ```markdown
     <!-- [TABLE] 4-column financial data -->  
     <!-- [CODE] Unidentified language (lines 12-25) -->  
     ```  

3. **Noise Filtering**  
   - Auto-ignore repeating footers/headers (e.g., "Page 3 of 23").  
   - If uncertain, keep content but flag: `<!-- CHECK: Possible footer -->`.  

#### **Phase 2: Strict 1:1 Conversion**  
1. **Text Formatting**  
   - Bold/italics → `**text**`/`*text*`  
   - Lists: Maintain exact indentation (even if inconsistent in PDF).  

2. **Tables**  
   - Convert **all rows** - never truncate.  
   - Use pipe syntax with alignment hints:  
     ```markdown
     | Header 1 | Header 2 |  
     |----------|----------|  
     | Row 1    | Data     |  <!-- Preserve empty cells! -->  
     ```  

3. **Code Blocks**  
   - Minimum 3-line backtick fences with line breaks:  
     ````markdown  
     ```python  
     def example():  # Never join split code lines!  
         pass  
     ```  
     ````  
   - For unidentified languages:  
     ```  
     [UNKNOWN_LANGUAGE]  
     fn obscure_code() { ... }  
     ```  

4. **Images/Figures**  
   - Placeholder + filename: `![Fig.3: Architecture Diagram](pdf_image_7.png)`  

#### **Phase 3: Validation**  
1. **Anti-Summarization Checks**  
   - Compare word count of original PDF paragraphs vs. Markdown output.  
   - If >5% discrepancy, revert and flag: `<!-- LENGTH MISMATCH: p.14 paragraph 2 -->`.  

2. **Ambiguity Protocol**  
   - For unclear formatting:  
     ```markdown  
     <!-- RAW_PDF_EXTRACT_START -->  
     [Strange   spacing]  
     <!-- RAW_PDF_EXTRACT_END -->  
     ```  

#### **Phase 4: Delivery**  
- Output **raw Markdown only** (no JSON/XML wrappers).  
- Include this completion token: `<!-- CONVERSION_COMPLETE_VERBATIM -->`.  

**Failure Modes That Void Approval:**  
- Any summarized/joined paragraphs  
- Truncated tables or code  
- Unmarked language guesses
```