Update common/ResponseValidation.py

common/ResponseValidation.py

@@ -6,43 +6,34 @@ from RespondentAgent import *
 from langchain_groq import ChatGroq
 
 
-def is_first_person(answer, processor_llm):
-    prompt = f"""
-You are an expert in analyzing writing style and narrative perspective.
-Determine whether the following response is written from a first-person point of view. 
-A first-person response includes pronouns such as "I", "me", "my", "mine", "we", "our", or "us" and is written from the perspective of the speaker.
-Do not guess. Only say "Yes" if the writing is clearly in first person. Otherwise, say "No".
+def matches_user_speaking_style(answer, transcript_text, processor_llm, user_profile, agent_question):
+    """
+    Uses the LLM to determine if the answer matches the tone and style of the user's prior speaking style in the transcript.
+    Returns True if similar and in first person, False otherwise.
+    """
+    logging.info("[Style Match Check] Entry")
+    
+    try:
+        # First-person perspective check
+        fp_prompt = f"""
+You are an expert in analysing writing style and narrative perspective.
+Determine whether the following response is written from a first-person point of view.
+A first-person response includes pronouns like "I", "me", "my", "mine", "we", "our", or "us".
+Say "First Person: Yes" only if clearly first-person, else say "First Person: No".
+
 Response:
 \"\"\"{answer}\"\"\"
-Output strictly in the following format:
+Output format:
 First Person: Yes
 or
 First Person: No
 """
-    try:
-        response = processor_llm.invoke(prompt)
-        content = response.content.strip().lower()
-        if "first person: yes" in content:
-            return True
-        elif "first person: no" in content:
+        fp_response = processor_llm.invoke(fp_prompt)
+        fp_result = fp_response.content.strip().lower()
+        if "first person: no" in fp_result:
+            logging.warning("[Style Match Check] Failed first-person test")
             return False
-        else:
-            logging.warning(f"Unexpected output format from LLM for first person check: {content}")
-            return False
-    except Exception as e:
-        logging.error(f"LLM failed during first person check: {e}")
-        return False
-
 
-def matches_user_speaking_style(answer, transcript_text, processor_llm, user_profile, agent_question):
-    """
-    Uses the LLM to determine if the answer matches the tone and style of the user's prior speaking style in the transcript.
-    Returns True if similar, False otherwise.
-    Incorporates logic to skip style matching for factual questions and uses profile-based criteria.
-    """
-    logging.info("[Style Match Check] Entry")
-    
-    try:
         # Get communication profile
         style = user_profile.get_field("Communication", "Style")
         tone = user_profile.get_field("Communication", "Tone")
@@ -61,38 +52,39 @@ def matches_user_speaking_style(answer, transcript_text, processor_llm, user_pro
             logging.info("[Style Match Check] Question is factual — skipping style comparison")
             return True
 
+        # Style match prompt
         prompt = f"""
-        You are a writing style and tone analyst.
-        
-        Your job is to assess whether a new response sounds like it was written by the same person who spoke in the interview transcript — considering phrasing, vocabulary, tone, and sentence structure.
-        
-        ---
-        ### Prior Interview Transcript (how the user usually talks):
-        \"\"\"{transcript_text}\"\"\"
-        
-        ---
-        ### New Response:
-        \"\"\"{answer}\"\"\"
-        
-        ---
-        ### Style Profile Reference:
-        - Style: {style}
-        - Tone: {tone}
-        - Preferred Length: {length}
-        - Topics: {topics}
-        
-        ---
-        ### Instructions:
-        - Check if the *tone*, *style*, and *language* of the new response align with the transcript.
-        - Use the style profile for reference.
-        - Focus on phrasing, formality, sentence structure, expressiveness, and personal flair.
-        - Ignore topic similarity — you’re assessing delivery style.
-        - Reply only with one of the following:
-        
-        Style Match: Yes  
-        or  
-        Style Match: No
-                """
+You are a writing style and tone analyst.
+
+Your job is to assess whether a new response sounds like it was written by the same person who spoke in the interview transcript — considering phrasing, vocabulary, tone, and sentence structure.
+
+---
+### Prior Interview Transcript (how the user usually talks):
+\"\"\"{transcript_text}\"\"\"
+
+---
+### New Response:
+\"\"\"{answer}\"\"\"
+
+---
+### Style Profile Reference:
+- Style: {style}
+- Tone: {tone}
+- Preferred Length: {length}
+- Topics: {topics}
+
+---
+### Instructions:
+- Check if the *tone*, *style*, and *language* of the new response align with the transcript.
+- Use the style profile for reference.
+- Focus on phrasing, formality, sentence structure, expressiveness, and personal flair.
+- Ignore topic similarity — you’re assessing delivery style.
+- Reply only with one of the following:
+
+Style Match: Yes  
+or  
+Style Match: No
+"""
 
         logging.info("[Style Match Check] Invoking LLM with style comparison prompt")
         response = processor_llm.invoke(prompt)
@@ -105,23 +97,24 @@ def matches_user_speaking_style(answer, transcript_text, processor_llm, user_pro
             logging.info("[Style Match Check] Style mismatch detected")
             return False
         else:
-            logging.warning(f"[Style Match Check] Unexpected response format: {result}")
+            logging.warning(f"[Style Match Check] Unexpected output format: {result}")
             return False
 
     except Exception as e:
         logging.error(f"[Style Match Check] LLM failed during comparison: {e}")
         return False
 
+
 def validate_response(question, answer, user_profile_str, fast_facts_str, interview_transcript_text, respondent_type, ai_evaluator_agent, processor_llm):
     llm_mode_prompt = f"""
-    You are an expert in market research interview analysis. Given the following question, determine if it is:
-    - Exploratory: subjective, open-ended, opinion-based, or reflective (e.g., feelings, motivations, preferences, aspirations, values, beliefs, etc.)
-    - Fact-based: objective, factual, or directly verifiable from the respondent's profile or transcript (e.g., age, location, occupation, education, etc.)
-    Respondent Type: {respondent_type}
-    Question: {question}
-    Output strictly in this format:
-    Evaluation Mode: <Exploratory or Fact-based>
-    """
+You are an expert in market research interview analysis. Given the following question, determine if it is:
+- Exploratory: subjective, open-ended, opinion-based, or reflective (e.g., feelings, motivations, preferences, aspirations, values, beliefs, etc.)
+- Fact-based: objective, factual, or directly verifiable from the respondent's profile or transcript (e.g., age, location, occupation, education, etc.)
+Respondent Type: {respondent_type}
+Question: {question}
+Output strictly in this format:
+Evaluation Mode: <Exploratory or Fact-based>
+"""
     response = processor_llm.invoke(llm_mode_prompt)
     output = response.content.strip()
     evaluation_mode = "exploratory"
@@ -136,20 +129,20 @@ def validate_response(question, answer, user_profile_str, fast_facts_str, interv
 
     if evaluation_mode == "exploratory":
         eval_prompt = f"""
-        You are an expert market research evaluator. Given the following:
-        - User Profile: {user_profile_str}
-        - Fast Facts: {fast_facts_str}
-        - Interview Transcript: {interview_transcript_text}
-        - Respondent Type: {respondent_type}
-        - Question: {question}
-        - Answer: {answer}
-        Please rate the answer on a scale of 0-10 for:
-        1. Plausibility (how realistic, authentic, and in-character the response is, given the profile and context)
-        2. Relevance (how directly and completely the answer addresses the question)
-        Output strictly in this format:
-        Plausibility Rating: <0-10>
-        Relevance Rating: <0-10>
-        """
+You are an expert market research evaluator. Given the following:
+- User Profile: {user_profile_str}
+- Fast Facts: {fast_facts_str}
+- Interview Transcript: {interview_transcript_text}
+- Respondent Type: {respondent_type}
+- Question: {question}
+- Answer: {answer}
+Please rate the answer on a scale of 0–10 for:
+1. Plausibility (how realistic, authentic, and in-character the response is, given the profile and context)
+2. Relevance (how directly and completely the answer addresses the question)
+Output strictly in this format:
+Plausibility Rating: <0-10>
+Relevance Rating: <0-10>
+"""
         eval_response = processor_llm.invoke(eval_prompt)
         eval_text = eval_response.content.strip()
         plausibility = None
@@ -167,27 +160,24 @@ def validate_response(question, answer, user_profile_str, fast_facts_str, interv
                     logging.error(f"Error parsing relevance rating: {e}")
         logging.info(f"Exploratory evaluation: plausibility={plausibility}, relevance={relevance}")
         if plausibility is not None and relevance is not None:
-            if plausibility >= 8.0 and relevance >= 8.0:
-                if not is_first_person(answer, processor_llm):
-                    logging.warning("Did not pass style due to 3rd person use")
-                    return False
-                return True
+            return plausibility >= 8.0 and relevance >= 8.0
         return False
+
     else:
         logging.info("Performing fact-based evaluation (accuracy)...")
         eval_prompt = f"""
-        You are an expert market research evaluator. Given the following:
-        - User Profile: {user_profile_str}
-        - Fast Facts: {fast_facts_str}
-        - Interview Transcript: {interview_transcript_text}
-        - Respondent Type: {respondent_type}
-        - Question: {question}
-        - Answer: {answer}
-        Please rate the answer on a scale of 0-10 for:
-        1. Accuracy (how well the answer matches the facts in the profile, transcript, or fast facts; penalise any unsupported or fabricated content)
-        Output strictly in this format:
-        Accuracy Rating: <0-10>
-        """
+You are an expert market research evaluator. Given the following:
+- User Profile: {user_profile_str}
+- Fast Facts: {fast_facts_str}
+- Interview Transcript: {interview_transcript_text}
+- Respondent Type: {respondent_type}
+- Question: {question}
+- Answer: {answer}
+Please rate the answer on a scale of 0–10 for:
+1. Accuracy (how well the answer matches the facts in the profile, transcript, or fast facts; penalise any unsupported or fabricated content)
+Output strictly in this format:
+Accuracy Rating: <0-10>
+"""
         eval_response = processor_llm.invoke(eval_prompt)
         eval_text = eval_response.content.strip()
         accuracy = None
@@ -199,9 +189,26 @@ def validate_response(question, answer, user_profile_str, fast_facts_str, interv
                     logging.error(f"Error parsing accuracy rating: {e}")
         logging.info(f"Fact-based evaluation: accuracy={accuracy}")
         if accuracy is not None:
-            if accuracy >= 8.0:
-                if not is_first_person(answer, processor_llm):
-                    logging.warning("Did not pass style due to 3rd person use")
-                    return False
-                return True
+            return accuracy >= 8.0
+        return False
+
+
+def validate_styled_answer(agent_name, agent_question, styled_answer, user_profile, processor_llm, interview_transcript_text):
+    """
+    Validates whether the styled answer matches the user's typical speaking style using prior interview transcript and communication profile.
+    Returns True if stylistically aligned, False otherwise.
+    """
+    logging.info("[validate_styled_answer] Entry")
+    try:
+        is_valid = matches_user_speaking_style(
+            answer=styled_answer,
+            transcript_text=interview_transcript_text,
+            processor_llm=processor_llm,
+            user_profile=user_profile,
+            agent_question=agent_question
+        )
+        logging.info(f"[validate_styled_answer] Style validation result: {is_valid}")
+        return is_valid
+    except Exception as e:
+        logging.exception("[validate_styled_answer] Exception during style validation")
         return False