Update app.py

app.py

@@ -1,4 +1,5 @@
 from flask import Flask, render_template, request, jsonify, send_file
+import ollama
 import json
 import re
 from duckduckgo_search import DDGS
@@ -9,22 +10,16 @@ import urllib3
 import pandas as pd
 import io
 import ast
-from groq import Groq
-import os
 
 
 app = Flask(__name__)
 
 search_prompt = """
 The user will provide a detailed description of a technical problem they are trying to solve in the context of intellectual property (IP) and patents. Your task is to generate some (2 to 5) highly specific and relevant search queries for Google, aimed at finding research papers closely related to the user's problem. Each search query should:
-
 1. Be crafted to find research papers, articles, or academic resources that address similar issues or solutions.
 2. Be focused and precise, avoiding generic or overly broad terms.
-
 Provide the search queries in the following **JSON format**. There should be no extra text, only the search queries as values.
-
 **Example Output:**
-
 ```json
 {
     "1": "user authentication 5G cryptographic keys identity management",
@@ -34,15 +29,10 @@ Provide the search queries in the following **JSON format**. There should be no
 """
 
 infringement_prompt = """You are an expert assistant designed to evaluate the novelty and inventiveness of patents by comparing them with existing documents. Your task is to analyze the background of a given patent and the first page of a related document to determine how well the document covers the problems mentioned in the patent.
-
 # Instructions:
-
 Understand the Patent Background: Carefully read and comprehend the background information provided for the patent. Identify the key problems that the patent aims to address.
-
 Analyze the Document: Review the provided document. Focus on identifying any problems that are similar to those mentioned in the patent background.
-
 Evaluate Coverage: Assess how well the document covers the problems mentioned in the patent. Use the following scoring system:
-
 Score 5: The document explicitly discusses the same problems as the patent, indicating that the problems are not novel.
 Score 4: The document discusses problems that are very similar to those in the patent, significantly impacting the novelty of the patent's problems.
 Score 3: The document mentions problems that are somewhat similar to those in the patent, but the coverage is not extensive enough to fully block the novelty of the patent's problems.
@@ -50,12 +40,9 @@ Score 2: The document mentions problems that are similar in some ways but are cl
 Score 1: The document touches upon related problems but does not directly address the specific problems mentioned in the patent.
 Score 0: The document does not discuss any problems related to those in the patent.
 Provide a Score: Based on your analysis, provide a score from 0 to 5 indicating how well the document covers the problems mentioned in the patent.
-
 Justify Your Score: Briefly explain the reasoning behind your score, highlighting specific similarities or differences between the problems discussed in the patent and the document.
-
 # Output Format:
 No details or explanations are required, just the results in the required **JSON** format with no additional word.
-
 {
     'score': [Your Score],
     'justification': "[Your Justification]"
@@ -63,15 +50,12 @@ No details or explanations are required, just the results in the required **JSON
 """
 
 insight_prompt = """Analyze the technical document and extract key insights that could enhance the patent problem. Focus on identifying security vulnerabilities, technical problems, innovative technologies, research questions, and protocols mentioned in the document.
-
 # Instructions:
 1. Identify 3-5 key insights from the document that could enhance or inform the patent problem.
 2. Each insight should be concise (max 10 words) but informative.
 3. Focus on technical elements that could be valuable for improving the patent.
-
 # Output Format:
 Return only a JSON object with insights as an array, with no additional text:
-
 {
     "insights": [
         "Security vulnerability in authentication handshake",
@@ -84,7 +68,6 @@ Return only a JSON object with insights as an array, with no additional text:
 """
 
 refine_problem_prompt = """You are an expert consultant specializing in enhancing and refining technical problems to make them more sophisticated, novel, and inventive. Your task is to transform an initial technical problem description into two improved versions by integrating selected insights from technical documents.
-
 # Instructions:
 1. Carefully review the initial technical problem.
 2. Consider the selected insights from technical documents that could enhance the problem formulation.
@@ -96,7 +79,6 @@ refine_problem_prompt = """You are an expert consultant specializing in enhancin
    - Add appropriate constraints and technical requirements
    - Maintain coherence and technical feasibility
    - Suggest innovative directions without fully solving the problem
-
 # Output Format:
 Return the results in **strictly valid JSON** format with no additional text. 
 All string values (especially multiline descriptions) must:
@@ -117,31 +99,6 @@ Reference json:
 }
 """
 
-ai_select_prompt = """You are an expert in patent analysis and technology innovation. Your task is to select the most relevant insights from a list that would help enhance the patentability of an invention.
-
-INSTRUCTIONS:
-
-1. Review the patent background description carefully.
-
-2. Analyze each insight in the provided list and determine its relevance to the patent background.
-
-3. Select ONLY THE MOST RELEVANT insights that are:
-   - Directly relevant to the patent's technical area
-   - Provide unique perspectives or identify novel challenges
-   - Could significantly enhance the patentability or scope of the invention
-   - Are technically substantive and not obvious
-
-4. Ignore insights that are generic, obvious, or would weaken the patent case.
-
-5. SELECT A MAXIMUM OF 7 INSIGHTS, even fewer if only a few are truly relevant.
-
-6. Return ONLY the selected insights in a valid JSON array.
-
-OUTPUT FORMAT:
-Return only a valid JSON array of strings, each containing the exact text of a selected insight.
-Example: ["Insight 1", "Insight 3", "Insight 5"]"""
-
-
 def ask_llm(user_message, model='llama-3.3-70b-versatile', system_prompt="You are a helpful assistant."):
     client = Groq(api_key=os.environ.get("GROQ_API_KEY"))
 
@@ -456,10 +413,8 @@ def refine_problem():
         # Prepare the message for the LLM
         message = f"""Initial Technical Problem:
 {original_problem}
-
 Selected Insights:
 {', '.join(insights)}
-
 User Comments/Suggestions:
 {user_comments}
 """
@@ -549,7 +504,6 @@ def export_excel():
         print(f"Error exporting Excel: {e}")
         return jsonify({'error': str(e)})
 
-
 @app.route('/ai-select-insights', methods=['POST'])
 def ai_select_insights():
     data = request.json
@@ -563,24 +517,45 @@ def ai_select_insights():
         if not insights or len(insights) == 0:
             return jsonify({'error': 'No insights provided', 'selected_insights': []})
             
+        # Prepare the prompt for the LLM
+        ai_select_prompt = """You are an expert in patent analysis and technology innovation. Your task is to select the most relevant insights from a list that would help enhance the patentability of an invention.
+
+INSTRUCTIONS:
+
+1. Review the patent background description carefully.
+
+2. Analyze each insight in the provided list and determine its relevance to the patent background.
+
+3. Select ONLY THE MOST RELEVANT insights that are:
+   - Directly relevant to the patent's technical area
+   - Provide unique perspectives or identify novel challenges
+   - Could significantly enhance the patentability or scope of the invention
+   - Are technically substantive and not obvious
+
+4. Ignore insights that are generic, obvious, or would weaken the patent case.
+
+5. SELECT A MAXIMUM OF 7 INSIGHTS, even fewer if only a few are truly relevant.
+
+6. Return ONLY the selected insights in a valid JSON array.
+
+OUTPUT FORMAT:
+Return only a valid JSON array of strings, each containing the exact text of a selected insight.
+Example: ["Insight 1", "Insight 3", "Insight 5"]"""
+        
         # Prepare message content
         message = f"""PATENT BACKGROUND:
 {patent_background}
 
 ALL AVAILABLE INSIGHTS:
-{json.dumps(insights)}"""
-        
-        # Call Ollama with the message and prompt
-        ai_reply = ask_llm(
-            user_message=message,
-            system_prompt=ai_select_prompt
-        )
-        
+{json.dumps(insights)}"""      
+    
+        ai_reply = ask_llm(message, system_prompt=ai_select_prompt)
         print(f"AI SELECTION RESPONSE: {ai_reply}")
         
         # Parse the JSON array from the response
         # First strip markdown code blocks if present
         cleaned_reply = ai_reply.replace('```json', '').replace('```', '').strip()
+        
         try:
             selected_insights = json.loads(cleaned_reply)
             
@@ -602,4 +577,4 @@ ALL AVAILABLE INSIGHTS:
         return jsonify({'error': str(e), 'selected_insights': []})
 
 if __name__ == '__main__':
-    app.run(host="0.0.0.0", port=7860)
+    app.run(debug=True)