Update app.py

app.py

@@ -87,7 +87,7 @@ def get_questions(prompt, input_text, num_questions=3, max_retries=10):
 
     return new_questions
 
-async def generate_question_audio(question, voice="en-US-AriaNeural"):
+async def generate_question_audio(question, voice="en-US-JennyNeural"):
     clean_question = re.sub(r'[^A-Za-z0-9.,?! ]+', '', question)
     tts = edge_tts.Communicate(text=clean_question, voice=voice)
     with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file:
@@ -168,7 +168,7 @@ for param in HR_PARAMETERS_CONFIG:
 # SUmmary of improvement(function)
 
 def generate_improvement_suggestions():
-    model = genai.GenerativeModel('gemini-2.0-flash-lite')
+    model = genai.GenerativeModel('gemini-2.0-flash')
     difficulty_level = st.session_state.get("difficulty_level_select", "Beginner")
     level_string = difficulty_level.lower()
 
@@ -247,7 +247,7 @@ def generate_improvement_suggestions():
 
 
 def evaluate_answers():
-    model = genai.GenerativeModel('gemini-2.0-flash-lite')
+    model = genai.GenerativeModel('gemini-2.0-flash')
     # difficulty_level_select is the key for the difficulty selectbox in your sidebar
     difficulty_level = st.session_state.get("difficulty_level_select", "Beginner") 
     level_string = difficulty_level.lower()
@@ -399,41 +399,86 @@ def evaluate_answers():
         base_assessment_criteria_qualitative_non_hr = """
         For the OVERALL qualitative summary, assess responses based on:
         - Conceptual Understanding (effort and relevance more than perfect accuracy for the level)
-        - Communication Clarity (can the core idea be understood?)
+        - Communication Clarity (is the idea understandable and logically stated?)
         - Depth of Explanation (relative to expected level)
-        - Use of Examples (if any, and if appropriate for the level)
-        - Logical Flow (is there a basic structure or train of thought?)
+        - Use of Examples (if any, and appropriate for the level)
+        - Logical Flow (basic structure or reasoning flow)
+        
+        Focus on both understanding and reasoning. Responses should demonstrate thinking, not memorization.
         """
         per_question_scoring_guidelines_non_hr = f"""
-        For EACH question and its answer, provide a score from 0 to 5 points.
+        For EACH question and its answer, assign a score from 0 to 5 points.
         The candidate is at a {level_string} level.
-        Consider the following when assigning the per-question score:
-        - Effort: Did the candidate attempt a meaningful answer, even if partially incorrect?
-        - Relevance: Is the response at least partially related to the question topic?
-        - Clarity of thought for the candidate's level.
-        - Basic logical structure.
-        - Use of examples, if any were given and appropriate.
+        Use the numeric scale and notes below for calibration.
+        **Scoring Scale (per question):**
+        - **5 (Excellent / 90–100%)** → Comprehensive, accurate, and well-structured. Includes reasoning or an example. Rare and well-deserved.
+        - **4 (Good / 75–89%)** → Mostly correct, relevant, and clear. Minor conceptual gaps but good structure.
+        - **3 (Fair / 60–74%)** → Partially correct or lacks depth, but shows understanding and effort.
+        - **2 (Basic / 45–59%)** → One-line or short answer with minimal reasoning; incomplete or overly generic.
+        - **1 (Poor / 30–44%)** → Attempted but largely irrelevant or unclear.
+        - **0 (No Effort / <30%)** → Incorrect, off-topic, or explicitly “I don’t know”.
+        **Important Rules:**
+        - *One-word or one-line answers* (e.g., just definitions or keywords) must NOT score more than **2 out of 5**, regardless of correctness, because they lack reasoning and depth.
+        - Encourage clarity, structure, and explanation over memorized phrases.
         """
+        scoring_tightness_guidelines = """
+        **Scoring Calibration (Strictness Guidance):**
+        - Maintain a slightly tight scoring approach.
+        - Incomplete or short one-line answers score **below 60% (1–2 out of 5)**.
+        - Scores of **5/5 (100%)** should be **rare** — reserved for comprehensive, insightful, and well-reasoned answers.
+        - Most competent answers should fall between **3 and 4**.
+        - When unsure, choose the **lower score** to maintain scoring consistency.
+        """
+        
         if level_string == "beginner":
             level_specific_instructions_non_hr = """
-            You are an **extremely understanding, encouraging, and supportive** interview evaluator for a **BEGINNER/FRESHER**. Your primary goal is to **build confidence**.
-            **Scoring Guidelines for Beginners (0-5 points per question):**
-            - **5 points:** Accurate, clear, and well-structured answer. Shows clear effort and basic understanding.
-            - **4 points:** Mostly correct with minor gaps or unclear phrasing.Good attempt, relevant, shows some understanding or key terms (e.g., one/two relevant words).
-            - **3 points:** Partially correct with evident effort, but lacks clarity or completeness.
-            - **1-2 points:** Minimal effort, mostly irrelevant, but an attempt beyond silence.
-            - **0 points:** Candidate explicitly says "I don’t know", "I'm not sure", or provides placeholder/non-answers. No relevant effort or understanding shown.Incorrect or unrelated.
-            Provide VERY positive feedback.
+            You are a **supportive, understanding evaluator** for a **BEGINNER/FRESHER**.
+            Focus on clarity, effort, and attempt — not perfection.
+            Encourage learning through feedback, but ensure fair scoring.
+            **Scoring Guidelines (0–5):**
+            - **5** → Accurate, clear, well-structured, and shows strong effort and reasoning. Rare.
+            - **4** → Mostly correct, relevant, and shows basic reasoning or understanding.
+            - **3** → Partial correctness with effort; may lack completeness or flow.
+            - **1–2** → One-line or definition-only answers; minimal reasoning. (Below 60%)
+            - **0** → No effort or irrelevant response.
+            Avoid giving high scores to short, memorized, or definition-only responses.
+            Provide motivating feedback that highlights areas of improvement.
             """
         elif level_string == "intermediate":
-            level_specific_instructions_non_hr = """Supportive evaluator for **INTERMEDIATE**. Scoring (0-5): 5=Correct/Clear; 3-4=Mostly correct; 1-2=Partial/Gaps; 0=Incorrect."""
+            level_specific_instructions_non_hr = """
+            You are a **balanced and fair evaluator** for an **INTERMEDIATE** candidate.
+            Expect conceptual clarity, structured reasoning, and relevant examples.
+            Be encouraging yet objective in scoring.
+            **Scoring Guidelines (0–5):**
+            - **5** → Clear, accurate, structured response with reasoning and relevance. Rare.
+            - **4** → Mostly correct with some logical structure and explanation.
+            - **3** → Some understanding; missing clarity or key detail.
+            - **1–2** → Short, definition-like, or minimal response. (Below 60%)
+            - **0** → Irrelevant or incorrect.
+            Never assign high scores to one-line or superficial answers.
+            """
         else: # Advanced
-            level_specific_instructions_non_hr = """Discerning evaluator for **ADVANCED**. Scoring (0-5): 5=Accurate/Comprehensive; 3-4=Correct lacks nuance; 1-2=Inaccurate; 0=Fundamentally incorrect."""
+            level_specific_instructions_non_hr = """
+            You are a **discerning but fair evaluator** for an **ADVANCED** professional.
+            Expect precision, applied understanding, and structured reasoning.
+            Maintain fairness without excessive strictness.
+            **Scoring Guidelines (0–5):**
+            - **5** → Exceptionally comprehensive, insightful, and accurate. (Rare)
+            - **4** → Correct and well-reasoned; may lack minor nuance or application.
+            - **3** → Adequate but missing depth, structure, or examples.
+            - **1–2** → Generic, incomplete, or one-line responses without reasoning. (Below 60%)
+            - **0** → Fundamentally incorrect or irrelevant.
+            Be concise and consistent in judgment; reward depth, not brevity.
+            
+            """
         
         evaluation_prompt_template_non_hr = f"""
         {level_specific_instructions_non_hr} 
         {per_question_scoring_guidelines_non_hr}
         {base_assessment_criteria_qualitative_non_hr}
+        {scoring_tightness_guidelines}
+         When evaluating, be supportive yet fair. Encourage clarity and effort but avoid over-rewarding shallow or memorized answers.
+        Maintain a balanced tone — neither too strict nor too lenient.
         **YOUR RESPONSE MUST STRICTLY FOLLOW THIS FORMAT. PROVIDE SCORES FOR EACH QUESTION.**
         Output format:
         **Per-Question Scores:**
@@ -445,7 +490,9 @@ def evaluate_answers():
         - Depth of Explanation: [Overall qualitative feedback here]
         - Examples: [Overall qualitative feedback here]
         - Logical Flow: [Overall qualitative feedback here]
+        - Final Remarks: [Brief encouraging but fair closing note]
         [Any additional overall encouraging remarks can optionally follow here]
+        Provide the final tone as **professional, balanced, and confidence-building**.
         """
         candidate_responses_formatted_non_hr = "\n\n".join(
             [f"Question {i+1}: {entry['question']}\nAnswer {i+1}: {str(entry.get('response', '[No response provided]'))}" for i, entry in enumerate(st.session_state["answers"])]
@@ -516,57 +563,130 @@ def evaluate_answers():
 ########################################///////////////////////////////////////////////////#########################################
 # --- Prompts for Question Generation ---
 BEGINNER_PROMPT = """
-You are a friendly mock interview trainer conducting a **Beginner-level** spoken interview in the domain of **{domain}**.
-Ask basic verbal interview questions based on the candidate's input: **{input_text}**.
-Guidelines:
-- Ask simple conceptual questions.
-- Avoid jargon and complex examples.
-- Use easy language.
-- No coding or technical syntax required.
-Ensure the questions are clear, to the point, and suitable for a {difficulty_level}-level interview in {selected_domain}.
-**New Requirement:**
-🚫 **Do NOT repeat any questions from previous generations again and again.** Ensure all generated questions are unique and different from past sessions.
-**Guidelines:**
-    ✅ Questions should focus on key concepts, best practices, and problem-solving within {selected_domain}.
-    ✅ Ensure questions are direct, structured, and relevant to real-world applications.
-    ❌ Do NOT include greetings like 'Let's begin' or 'Welcome to the interview'.
-    ❌ Avoid vague or open-ended statements—each question should be concise and specific.
+You are a friendly mock interview trainer conducting a beginner-level spoken interview in SQL.
+
+Candidate background:
+{input_text}
+
+Interview scope (MANDATORY):
+- DBMS (RDBMS, types of RDBMS, use)
+- SQL Clauses (SELECT, WHERE, GROUP BY, HAVING, ORDER BY)
+- SQL Constraints (PRIMARY KEY, FOREIGN KEY, UNIQUE, NOT NULL, CHECK)
+- SQL Joins (INNER, LEFT, RIGHT, FULL – basics only)
+- SQL Views
+- SQL Commands (DDL, DML, DQL, DELETE vs TRUNCATE vs DROP)
+
+Question Rules (STRICT):
+1. Ask ONLY beginner-level conceptual questions.
+2. Avoid coding syntax, queries, or SQL keywords formatting.
+3. Use simple, spoken English—as if asking a fresher.
+4. Each question must focus on **one clear concept**.
+5. Questions must be **practical and real-world oriented**.
+
+
+- DO NOT rephrase, repeat, or slightly modify any of the above.
+- Generate **new and conceptually different questions only**.
+
+Cross-Questioning Rule:
+- If the candidate answers a concept (assume basic understanding),
+  ask ONE logical follow-up that tests clarity or application.
+- Follow-up must be related but NOT repetitive.
+
+Structure Requirement:
+- Ask 5 questions in total:
+   • 3 primary concept questions from different SQL areas
+   • 2 follow-up cross questions based on understanding
+
+❌ Do NOT include greetings or closing statements.
+❌ Do NOT ask vague questions like “Explain SQL” or “What is a database”.
+❌ Do NOT repeat topics already covered within the same session.
+
 """
 
 INTERMEDIATE_PROMPT = """
-You are a professional mock interviewer conducting an **Intermediate-level** spoken interview in the domain of **{domain}**.
-Ask moderately challenging verbal interview questions based on the candidate's input: **{input_text}**.
-Guidelines:
-- Use a mix of conceptual and real-world scenario questions.
-- Include light critical thinking.
-- Still no need for code, formulas, or complex diagrams.
-- No coding or technical syntax required.
-Ensure the questions are clear, to the point, and suitable for a {difficulty_level}-level interview in {selected_domain}.
-**New Requirement:**
-🚫 **Do NOT repeat any questions from previous generations again and again.** Ensure all generated questions are unique and different from past sessions.
-**Guidelines:**
-    ✅ Questions should focus on key concepts, best practices, and problem-solving within {selected_domain}.
-    ✅ Ensure questions are direct, structured, and relevant to real-world applications.
-    ❌ Do NOT include greetings like 'Let's begin' or 'Welcome to the interview'.
-    ❌ Avoid vague or open-ended statements—each question should be concise and specific.
+You are a professional mock interviewer conducting an intermediate-level spoken interview in the domain of {selected_domain}.
+
+Candidate background:
+{input_text}
+
+Interview Scope (MANDATORY):
+- Core concepts relevant to {selected_domain}
+- Best practices and common mistakes
+- Real-world usage scenarios
+- Decision-making and trade-off reasoning
+
+Question Design Rules (STRICT):
+1. Ask intermediate-level verbal questions that test understanding, not memory.
+2. Questions must combine:
+   • a concept
+   • a practical situation or decision
+3. No coding, formulas, syntax, or step-by-step instructions.
+4. Use clear, spoken English suitable for interview conversation.
+5. Each question must assess reasoning, explanation, or choice justification.
+
+
+- DO NOT repeat, rephrase, paraphrase, or partially overlap with these questions.
+- Each new question must introduce a **new angle or scenario**.
+
+Cross-Questioning Logic:
+- For at least 2 questions, ask a follow-up that:
+   • challenges the candidate’s reasoning
+   • explores consequences or alternatives
+- Follow-up questions must deepen evaluation, not repeat the concept.
+
+Structure Requirement:
+- Ask exactly 6 questions:
+   • 4 primary scenario-based questions
+   • 2 logically connected follow-up questions
+- Ensure questions span **different sub-topics** within {selected_domain}.
+
+❌ Do NOT include greetings, closing statements, or filler sentences.
+❌ Do NOT ask generic questions like “Explain” or “What is”.
+❌ Avoid theoretical definitions unless tied to a real-world scenario.
+
 """
 
 ADVANCED_PROMPT = """
-You are a strict mock interviewer conducting an **Advanced-level** spoken interview in the domain of **{domain}**.
-Ask deep, analytical, real-world scenario-based questions from the candidate's input: **{input_text}**.
-Guidelines:
-- Expect detailed, logical, well-structured answers.
-- Include challenging “why” and “how” based questions.
-- No need for code, but assume candidate has high expertise.
-- No coding or technical syntax required.
-Ensure the questions are clear, to the point, and suitable for a {difficulty_level}-level interview in {selected_domain}.
-**New Requirement:**
-🚫 **Do NOT repeat any questions from previous generations again and again.** Ensure all generated questions are unique and different from past sessions.
-**Guidelines:**
-    ✅ Questions should focus on key concepts, best practices, and problem-solving within {selected_domain}.
-    ✅ Ensure questions are direct, structured, and relevant to real-world applications.
-    ❌ Do NOT include greetings like 'Let's begin' or 'Welcome to the interview'.
-    ❌ Avoid vague or open-ended statements—each question should be concise and specific.
+You are a strict, senior-level mock interviewer conducting an advanced spoken interview in the domain of {selected_domain}.
+
+Candidate background:
+{input_text}
+
+Interview Objective:
+Evaluate deep expertise, judgment, trade-offs, and real-world decision-making ability within {selected_domain}.
+
+Question Design Rules (VERY STRICT):
+1. Ask advanced-level questions that assume strong foundational knowledge.
+2. Every question must involve:
+   • a real-world scenario, constraint, or failure case
+   • a clear “why” or “how” decision explanation
+3. No coding, formulas, or technical syntax.
+4. Questions must demand structured, multi-point reasoning.
+5. Each question should test impact, risk, scalability, or maintainability.
+
+🚫 Absolute Non-Repetition Rule:
+- The following questions have already been asked across previous sessions:
+
+- DO NOT repeat, paraphrase, restructure, or partially overlap with these questions.
+- Each new question must explore a **new risk, trade-off, or decision dimension**.
+
+Advanced Cross-Examination Logic:
+- For at least 3 questions, include a follow-up that:
+   • challenges assumptions
+   • introduces a constraint (scale, performance, data quality, security, cost)
+   • forces the candidate to defend or revise their approach
+- Follow-ups must **escalate difficulty**, not restate the concept.
+
+Structure Requirement:
+- Ask exactly 7 questions total:
+   • 4 primary deep-dive scenario questions
+   • 3 targeted follow-up cross-examination questions
+- Ensure coverage across **different high-impact areas** of {selected_domain}.
+
+❌ Do NOT include greetings, closings, or filler language.
+❌ Do NOT ask definition-based or descriptive questions.
+❌ Avoid broad “What would you do?”—each question must have a specific context.
+
 """
 
 ########################################///////////////////////////////////////////////////#########################################
@@ -722,7 +842,7 @@ if st.session_state["selected_domain"] == "Soft Skills":
                 st.warning("⚠️ Please upload a resume.")
                 st.stop()
             prompt = f"Ask {num_qs} HR-style interview questions based on this resume: {input_text}"
-            model = genai.GenerativeModel('gemini-2.0-flash-lite')
+            model = genai.GenerativeModel('gemini-2.0-flash')
             response = model.generate_content([prompt])
             questions = [q.strip("* ") for q in response.text.strip().split("\n") if q.strip()]
             st.session_state["generated_questions"] = questions[:num_qs]
@@ -863,7 +983,7 @@ else:
             st.session_state["generated_questions"] = sample(questions_from_excel, min(num_qs, len(questions_from_excel)))
         else:
             prompt = f"Ask {num_qs} direct and core-level {difficulty} interview questions related to {input_text}. Do not include intros or numbering."
-            model = genai.GenerativeModel('gemini-2.0-flash-lite')
+            model = genai.GenerativeModel('gemini-2.0-flash')
             response = model.generate_content([prompt, input_text])
             lines = response.text.strip().split("\n")
             questions = [q.strip("* ") for q in lines if q.strip()]