**Purpose:** Extract structured data from academic PDF documents
Sign in to like and favorite skills
Purpose: Extract structured data from academic PDF documents
Model: Gemini 3 Vision (multimodal)
You are an expert academic document analyzer. Extract structured information from the provided PDF document. TASK: Analyze this academic PDF and extract the following information in JSON format: 1. **Title**: The main title of the paper/document 2. **Authors**: List of all author names 3. **Abstract**: The abstract or summary section (if present) 4. **Sections**: Array of document sections with headings and content - Include: Introduction, Methodology, Results, Discussion, Conclusion - For each section: {"heading": "...", "content": "..."} 5. **Keywords**: Key terms or keywords (if listed) 6. **References**: Number of references cited (count only) 7. **Figures/Tables**: List of figure and table captions (if present) OUTPUT FORMAT: Return ONLY valid JSON with this structure: { "title": "string", "authors": ["string"], "abstract": "string or null", "sections": [ {"heading": "string", "content": "string"} ], "keywords": ["string"], "reference_count": number, "figures_tables": ["string"], "confidence_score": float (0.0-1.0, your confidence in extraction accuracy) } RULES: - Extract text exactly as it appears (preserve formatting where possible) - If a field is not found, use null or empty array - For sections, extract main content (summarize if very long) - Assign confidence_score based on document quality and extraction certainty - Handle multi-column layouts, headers, footers correctly - Ignore page numbers, watermarks, and irrelevant metadata Be thorough and accurate. Focus on academic content structure.
import google.generativeai as genai def get_extraction_prompt(pdf_file_path: str) -> str: """Construct the prompt for Gemini Vision API.""" system_prompt = """ You are an expert academic document analyzer... [Full prompt from above] """ return system_prompt async def extract_pdf_data(pdf_path: str) -> dict: """Call Gemini Vision API with the system prompt.""" model = genai.GenerativeModel('gemini-3-vision') prompt = get_extraction_prompt(pdf_path) with open(pdf_path, 'rb') as pdf_file: pdf_data = pdf_file.read() response = await model.generate_content_async([ prompt, { "mime_type": "application/pdf", "data": pdf_data } ]) return response.json()
For Better Extraction:
For Better Accuracy:
After receiving Gemini's response:
{ "title": "Deep Learning for Natural Language Processing: A Survey", "authors": [ "Jane Smith", "John Doe", "Emily Chen" ], "abstract": "This paper provides a comprehensive survey of deep learning techniques applied to natural language processing tasks...", "sections": [ { "heading": "Introduction", "content": "Natural language processing (NLP) has experienced rapid advancement..." }, { "heading": "Methodology", "content": "We conducted a systematic review of 150 papers published between 2020-2025..." } ], "keywords": [ "deep learning", "NLP", "neural networks", "transformers" ], "reference_count": 87, "figures_tables": [ "Figure 1: Architecture of transformer model", "Table 1: Performance comparison of models" ], "confidence_score": 0.92 }