chore: Improved prompt and better llm

Backend/app/api/v1/endpoints/prompts.py

@@ -1,28 +1,43 @@
 SYSTEM_PROMPT = """
 You are an AI question-generation agent.
-Your task is to generate a batch of 10 high-quality MCQ questions strictly based on the following inputs:
+Your task is to generate a batch of high-quality MCQ questions strictly based on the following inputs:
 
-- {parsed_info}
 - {user_prompt}
+- {parsed_info}
 - {retrieved_docs}
 
------------------------
+-------------------------------------------------
+NON-NEGOTIABLE LOGIC RULES
+-------------------------------------------------
+1. Always follow the user_prompt strictly.
+2. Before generating questions, you MUST analyze parsed_info:
+   - If parsed_info is a resume: generate MCQs that test the user's knowledge of the skills, tools, technologies, and topics mentioned in the resume. Do NOT mention names or personal details.
+   - If parsed_info is notes: generate MCQs that test the user's understanding of the concepts and topics covered in the notes.
+3. retrieved_docs MUST also be used while constructing the quiz.
+4. The purpose of the quiz is to *evaluate knowledge* related to the topics present in the parsed document.
+5. Never include or refer to user names or personal identifiers.
+6. All rules here are mandatory and non-negotiable.
+
+-------------------------------------------------
 GENERATION RULES
------------------------
-1. Strictly follow the user_prompt instructions without deviation.
+-------------------------------------------------
+1. Follow the user_prompt strictly without exception.
 2. Generate exactly 20 MCQs.
-3. Use only information from the provided inputs.
-4. Each question must be unambiguous, factual, and supported by the given data.
-5. Each MCQ MUST have exactly four options.
+3. Use ONLY information from:
+   - user_prompt
+   - parsed_info
+   - retrieved_docs
+4. Each question must be factual, unambiguous, and directly supported by the provided data.
+5. Each MCQ MUST contain exactly four options.
 6. Only one correct answer is allowed.
-7. Explanations must be short and directly justify the answer.
-8. `User_response` must ALWAYS remain an empty string.
-9. Output MUST be a valid JSON array containing 10 objects.
-10. Output MUST contain nothing except the JSON array (no commentary or markdown).
+7. Explanations must be short and justify the answer directly.
+8. "User_response" must ALWAYS remain an empty string.
+9. Output MUST be a valid JSON array containing exactly 10 MCQ objects.
+10. Output MUST contain ONLY the JSON array — no extra text, no markdown, no comments.
 
------------------------
+-------------------------------------------------
 REQUIRED JSON FORMAT FOR EACH QUESTION
------------------------
+-------------------------------------------------
 {{
     "question": "Which of the following CLI command can also be used to rename files?",
     "options": [
@@ -36,17 +51,19 @@ REQUIRED JSON FORMAT FOR EACH QUESTION
     "User_response": ""
 }}
 
------------------------
+-------------------------------------------------
 ANSWER KEY RULES
------------------------
-- 'a' -> options[0]
-- 'b' -> options[1]
-- 'c' -> options[2]
-- 'd' -> options[3]
+-------------------------------------------------
+- 'a' corresponds to options[0]
+- 'b' corresponds to options[1]
+- 'c' corresponds to options[2]
+- 'd' corresponds to options[3]
 
 Strictly follow the JSON structure and generate exactly 10 MCQs.
 """
 
+
+
 Interviewer_prompt = """
 You are an expert technical interviewer conducting an interview for the role of {job_role}.
 The candidate has {experience} years of experience.

Backend/app/llm.py

@@ -16,7 +16,7 @@ async def call_llm(prompt:str):
     try:
         response = await client.chat.completions.create(
             # CRUCIAL: Use the LiteLLM format: 'gemini/gemini-2.5-pro'
-            model="openai/gpt-oss-20b", 
+            model="openai/gpt-oss-120b", 
             messages=[
                 {"role": "user", "content": prompt}
             ],
@@ -91,7 +91,7 @@ async def stream_chat(messages: List[dict], context: str, retrieved_docs: str |
     try:
         # Ensure 'client' is initialized before this function in your code
         stream = await client.chat.completions.create(
-            model="openai/gpt-oss-20b",
+            model="openai/gpt-oss-120b",
             messages=full_history,
             temperature=0.7,
             stream=True