Scientific Literature → JSON Extractor

Wajahat Hussain

1) Choose input type

Input mode
PDF file

No PDF loaded.

Preview extracted text (first ~2000 chars) 

      
Preview detected abstract
        

    


    
  


  

    
2) Extraction setup

    

      

        Mode
        
      

      

        Provider
        
      

      

        Model (optional)
        
      

    


    


    
3) JSON Schema

    

    

    

      
      
      
      
      
    


    

  


  

    
Result (JSON)

    
{}

  


  

    
Cross-Model Review

    
Runs review against the current extraction JSON using the other 3 providers.

    

      

        
Reviewer 1

        

          
Correct
-

          
Incorrect
-

        

        
Missing properties (JSON)
[]

        
Notes
None

        
Raw / error output

      


      

        
Reviewer 2

        

          
Correct
-

          
Incorrect
-

        

        
Missing properties (JSON)
[]

        
Notes
None

        
Raw / error output

      


      

        
Reviewer 3

        

          
Correct
-

          
Incorrect
-

        

        
Missing properties (JSON)
[]

        
Notes
None

        
Raw / error output

      

    

  


  
  

    
Q&A & Logic Pipeline (closed-book LLM vs Prolog over the JSON KB)

    

      Each provider proposes 2 questions about the paper (sequentially, with dedup), giving 8 total.
      For each question:
      Path A — closed-book LLM with access only to the JSON KB.
      Path B — natural-language question is rewritten to a Prolog query, which is then
      actually executed by tau-prolog against the JSON-derived fact base. Real bindings are returned.
      The two paths are compared deterministically.
    


    

      

        Path A answer provider
        
      

    


    
Not run yet.

    

    


    

      Generated Prolog facts
      

    

    

      Raw pipeline JSON output