utils/prompt_templates/page_templates/generate_baseline_full_content.yaml

system_prompt: |
  You are a helpful academic expert and web developer, who is specialized in generating a paper project page, from given research paper's contents and figures.

template: |
  Below is the raw content with markdown text, images, and tables information:
  <raw_content>
  {{raw_content}}
  </raw_content>

  Your task is to analyze the paper content and generate a complete structured full_content JSON that contains ALL the information needed for the final HTML webpage. This JSON will be the single source of truth for generating the project page.
  
  You need to:
  
  1. **Extract Paper Metadata**:
     - Paper title
     - Authors with their affiliations (use <sup> tags for superscripts)
     - Institution affiliations
     - Any other relevant metadata (conference, year, links, etc.)
  
  2. **Analyze and Plan Paper Sections**:
     - Identify the main sections from the paper (Abstract, Introduction, Method, Results, Analysis, Conclusion, etc.)
     - For each section, extract the key content that should appear on the project page
     - Write clear, concise content summaries that will be displayed
     - DO NOT just copy raw paper text - synthesize and adapt it for web presentation
  
  3. **Select and Place Visual Elements**:
     - Identify the teaser figure (the most important visualization, usually appears first)
     - For each section, select the most relevant figures and tables
     - Use the EXACT paths provided in raw_content for all images and tables
     - Include the exact width and height values from raw_content
     - Write descriptive captions for each visual element
     - Each figure/table should be used at most once
     - Ensure all important figures are included
     - For sections with multiple tables, choose only the most relevant one
  
  4. **Content Guidelines**:
     - The teaser figure must be included and should appear early (typically in Overview or after Abstract)
     - Prioritize pictures and tables based on their relevance and importance
     - Ensure figures are closely related to their section's content
     - Match visual elements with their corresponding text discussions
     - Specify clear placement instructions for each visual element
     - Write content that flows naturally and is appropriate for a web page (not raw academic text)
  
  5. **Path and Dimension Requirements**:
     - Use EXACTLY the same paths as provided in raw_content (e.g., "assets/paper-picture-8.png")
     - Include the exact width and height values from raw_content
     - Maintain the original aspect ratios of all visual elements
  
  Please provide your complete full_content structure in the following JSON format:
  
  ```json
  {
    "title": "Complete paper title",
    "authors": "Author names with <sup> tags for affiliations, e.g., 'John Doe<sup>1</sup>, Jane Smith<sup>2</sup>*'",
    "affiliation": "Complete affiliation text with <sup> tags, e.g., '<sup>1</sup>MIT, <sup>2</sup>Stanford University'",
    "teaser_figure": {
      "path": "exact path from raw_content",
      "description": "detailed description of the teaser figure",
      "width": "width value from raw_content",
      "height": "height value from raw_content",
      "caption": "caption text for the teaser"
    },
    "Section Name 1": "Complete content text for this section. This should be well-written, web-appropriate content that synthesizes the paper's key points. Include inline references to figures like: [Figure description][path][width=X, height=Y](figure_number) when you want to reference a visual element.\n\n![Detailed caption describing what the figure shows][assets/exact-path.png][width=1234, height=567](1)",
    "Section Name 2": "Content for the next section with its own flow and structure...\n\n![Another figure caption][assets/another-path.png][width=890, height=456](2)",
    "Section Name 3": "More content...\n\n![Table caption][assets/table-path.png][width=2000, height=800](3)"
  }
  ```
  
  CRITICAL Requirements for the JSON structure:
  
  1. **Metadata Fields** (required at the top):
     - "title": The full paper title
     - "authors": Author names with superscript affiliations
     - "affiliation": Institution information with superscripts
     - "teaser_figure": A separate object with path, description, width, height, and caption
  
  2. **Section Fields** (one per major paper section):
     - Use clear section names as keys (e.g., "Overview", "Method", "Experimental Results")
     - Each section's value should be a string containing:
       * Well-written, web-appropriate content that explains the section
       * Embedded figure/table references using the notation: ![caption][path][width=X, height=Y](number)
       * The figure notation MUST be on a new line (with \n\n before it)
       * Natural flow and transitions between content and figures
  
  3. **Figure/Table Notation Format**:
     - Use: ![Caption text][exact/path/from/raw_content][width=1234, height=567](figure_number)
     - The figure_number must be a unique integer (1, 2, 3, ...)
     - Caption should describe what the visual shows
     - Path must EXACTLY match raw_content
     - Width and height must EXACTLY match raw_content
     - Place figures after the relevant text that discusses them
  
  4. **Content Writing Guidelines**:
     - Write clear, engaging content suitable for a project page (not raw academic prose)
     - Each section should tell a coherent story
     - Ensure smooth transitions between text and visuals
     - Highlight key contributions and findings
     - Keep the tone professional but accessible
     - DO NOT just copy-paste from the paper - adapt and synthesize
  
  5. **Visual Placement Strategy**:
     - Teaser figure: Separate field, will be placed prominently at the top
     - Section figures: Embedded in section text where most relevant
     - Place figures after the text that introduces or discusses them
     - Ensure balanced distribution of visuals across sections
     - Don't overload any single section with too many visuals
  
  Important reminders:
  - All paths must EXACTLY match those in raw_content
  - All width and height values must EXACTLY match those in raw_content
  - Figure numbers should be sequential and unique across the entire document
  - Each visual element should appear only once
  - The teaser figure should be the most impactful/representative visualization
  - Section names should be clear and match the paper's structure
  - Content should be web-friendly, not just copied academic text
  - Use \n\n before figure notations to ensure they're on new lines in the JSON string

jinja_args:
  - raw_content