AI research prototype · Built by Stephen Lese

Interview Extraction and Claim Checking

Turn long interviews into findings you can trace to a source, compare across people and use to test a conclusion.

Open the demo

Enter the shared demo passcode to get started or return to your saved work.

The same passcode opens the Credit Waiver Agent demo. Kept only for this browser session.

Back to portfolio
This project changed in another tab. Reload to use the latest saved work.

Start with the Harbor brief

Explore a fictional payments diligence project using a sample brief and two expert interviews. Load the brief to see the decision and questions that will guide the AI.

Sample brief · Read only

Read the brief · Download brief

Nothing is analyzed until you select interviews and press Extract. This demo uses only the supplied Harbor materials.

This saved run has no interviews

Your saved brief and notes are preserved. Continue with the Harbor example to try the sample interviews.


      
    

    


      


      

        
Add a transcript

        

        
Choose one or both fictional Harbor interviews. You can read or download them first. Extract sends the selected samples to the AI; results are saved in this browser.

        
        

        

        


        

          
          
          
          
        

      


      

        

        


        

          
          
          
          
          
          
        


        

          

            Reviewing —. Each finding was proposed by the model and
            carries its source quotation. Quote located means the quoted words
            were found in the file. A separate model review checks the interpretation
            for brief-guided findings. Both checks can miss errors, so read the quote before confirming.
          

          

            
            
          

          

            0 of 0
            confirmed across the panel
            
            
          

          

        


        

          

            Compare confirmed findings across interviews. Each entry names its
            source. A gap means no finding has been confirmed for that question.
          

          

        


        

          

        


        

          

            Drafted from confirmed findings only, across the whole panel. Each drafted claim
            cites the findings it rests on and each finding carries its expert. A claim covering
            all experts needs support from all of them.
          

          

            
            
          

          

          


          
Your notes

          
          

        


        

          

            Write a conclusion you want to test. The AI checks it against confirmed findings
            and returns a verdict, supporting evidence, caveats and a source note you can reuse.
          

          

            
            

              
              
              
            

          

          

          

        


        

          

            The finished synthesis, assembled from what you confirmed and kept. Print it to
            PDF or copy it as markdown. Evidence includes source quotes and their locations.