Update modules/llm_handler.py

modules/llm_handler.py

@@ -6,37 +6,49 @@ from groq import Groq
 from modules.web_search import search_for_example_answers
 
 # Initialize the Groq client
-# It will automatically use the API key from your Hugging Face secrets
 client = Groq(api_key=os.environ.get("GROQ_API_KEY"))
-MODEL = "llama3-8b-8192" # Use Groq's Llama 3 8B model
+MODEL = "llama3-70b-8192" # Use the more powerful 70B model for detailed analysis
 
 def generate_question(interview_type, document_text):
     prompt = f"As an expert {interview_type} interviewer, ask one relevant, open-ended question based on this document:\n\n---\n{document_text}\n---"
     try:
         chat_completion = client.chat.completions.create(
-            messages=[{"role": "user", "content": prompt}],
-            model=MODEL,
+            messages=[{"role": "user", "content": prompt}], model=MODEL
         )
         return chat_completion.choices[0].message.content
     except Exception as e:
         return f"Error generating question from API: {e}"
 
 def evaluate_answer(question, answer):
+    # This new prompt demands a much higher level of detail
     prompt = f"""
-    You are an interview coach. Your task is to evaluate a candidate's answer.
-    You MUST provide a score from 1-10 for each of the following three categories.
-    The format MUST be exactly `Category: [SCORE]/10`.
+    You are a meticulous and insightful interview coach. Your task is to provide a highly detailed evaluation of a candidate's answer.
 
-    Factual Accuracy: [SCORE]/10
-    Relevance & Directness: [SCORE]/10
-    Structure & Clarity: [SCORE]/10
-    
-    After the scores, you MUST provide a brief written evaluation.
+    **INTERVIEW QUESTION:**
+    "{question}"
+    ---
+    **CANDIDATE'S ANSWER:**
+    "{answer}"
+    ---
+    **YOUR TASK:**
+    Provide a detailed, multi-part evaluation. You MUST include scores and written analysis for each section.
+
+    **1. Score Breakdown (Format: `Category: [SCORE]/10`)**
+    You must provide a score for each of the following:
+    - Factual Accuracy: [SCORE]/10
+    - Relevance & Directness: [SCORE]/10
+    - Structure & Clarity (STAR Method): [SCORE]/10
+
+    **2. Detailed Written Evaluation**
+    - **Strengths:** What did the candidate do well? (e.g., "Good use of a specific example...")
+    - **Areas for Improvement:** What were the key weaknesses? (e.g., "The result of the action was unclear...")
+
+    **3. Concrete Suggestion**
+    - **Example Rephrasing:** Provide a short example of how they could have phrased a key part of their answer more effectively.
     """
     try:
         chat_completion = client.chat.completions.create(
-            messages=[{"role": "user", "content": prompt}],
-            model=MODEL,
+            messages=[{"role": "user", "content": prompt}], model=MODEL
         )
         return chat_completion.choices[0].message.content
     except Exception as e:
@@ -46,28 +58,40 @@ def parse_scores_from_evaluation(evaluation_text: str) -> dict:
     scores = {
         'Factual Accuracy': 0,
         'Relevance & Directness': 0,
-        'Structure & Clarity': 0
+        'Structure & Clarity (STAR Method)': 0
     }
-    pattern = r"(Factual Accuracy|Relevance & Directness|Structure & Clarity):\s*\[?(\d{1,2})\]?\/10"
+    # Updated regex to match the new category name
+    pattern = r"(Factual Accuracy|Relevance & Directness|Structure & Clarity \(STAR Method\)):\s*\[?(\d{1,2})\]?\/10"
     matches = re.findall(pattern, evaluation_text, re.IGNORECASE)
     
     for match in matches:
         category_name, score_value = match[0].strip(), int(match[1])
-        if category_name in scores:
+        # Normalize the key for consistency
+        if "Structure" in category_name:
+            scores['Structure & Clarity (STAR Method)'] = score_value
+        elif category_name in scores:
             scores[category_name] = score_value
             
     print(f"📊 Parsed scores: {scores}")
     return scores
 
 def generate_holistic_feedback(full_interview_log):
+    # This prompt is also enhanced for more detail
     prompt = f"""
-    You are a senior interview coach. Based on the entire Q&A log, provide a high-level "Overall Performance Summary" and an "Actionable Improvement Plan".
-    **FULL INTERVIEW LOG:** --- {full_interview_log} ---
+    You are a senior career strategist reviewing a candidate's full interview performance.
+    Based on the entire Q&A log, provide a detailed "Overall Performance Summary" and an "Actionable Improvement Plan".
+
+    **FULL INTERVIEW LOG:**
+    ---
+    {full_interview_log}
+    ---
+    **YOUR TASK:**
+    1.  **Overall Performance Summary:** Write a detailed paragraph summarizing the candidate's performance. Analyze their communication style, confidence, and consistency. Identify the most significant recurring strengths and weaknesses across all answers.
+    2.  **Actionable Improvement Plan:** Provide a bulleted list of the top 3 most critical and specific actions the candidate must take. For each action, explain *why* it's important and provide a *concrete example*. (e.g., "- Action: Quantify your achievements. Why: It demonstrates impact. Example: Instead of 'improved the system,' say 'reduced server response time by 15%.'").
     """
     try:
         chat_completion = client.chat.completions.create(
-            messages=[{"role": "user", "content": prompt}],
-            model=MODEL,
+            messages=[{"role": "user", "content": prompt}], model=MODEL
         )
         return chat_completion.choices[0].message.content
     except Exception as e: