import os import json import time from groq import Groq from dotenv import load_dotenv from pydantic import BaseModel, Field, ValidationError from typing import List, Optional, Dict load_dotenv('.env.local') # Configure Groq API API_KEY = os.getenv("GROQ_API_KEY") client = None if API_KEY: client = Groq(api_key=API_KEY) # Pydantic models for robust validation class Question(BaseModel): question_text: str marks: Optional[int] = 0 bloom_level: Optional[str] = "Understand" type: Optional[str] = "Descriptive" options: Optional[List[str]] = None correct_answer: Optional[str] = None marking_scheme: Optional[str] = None estimated_time_mins: Optional[int] = None competency: Optional[str] = None chapter: Optional[str] = None internal_choice_question_text: Optional[str] = None class Section(BaseModel): section_name: str instructions: str questions: List[Question] class ExamPaper(BaseModel): title: str sections: List[Section] def generate_questions(content, configure_data, file_path=None, retry_count=2): """ Generates questions using Groq AI (Llama 3.3) with pydantic validation and retry logic. """ if not API_KEY or not client: return {"error": "Groq API Key not configured"} model_name = "llama-3.3-70b-versatile" system_prompt = "You are an expert academic examiner. Output ONLY valid JSON." # Format prompts (reused from original) blooms_dist = configure_data.get('bloomsDistribution', {}) question_types = configure_data.get('questionTypes', []) difficulty_score = configure_data.get('difficulty', 50) advanced = configure_data.get('advancedSettings', {}) blooms_str = ", ".join([f"{k.capitalize()} ({v}%)" for k, v in blooms_dist.items() if v > 0]) pattern_str = "" total_requested_questions = 0 num_sections = 0 for qt in question_types: if qt.get('selected'): count = int(qt.get('count', 0)) if count > 0: label = qt.get('label', 'Question') topic = qt.get('topic', '').strip() choices = int(qt.get('internalChoices', 0)) req = f"- Section for '{label}': EXACTLY {count} questions, {qt.get('marks', 0)} marks each." if topic: req += f" Focus on: '{topic}'." if choices > 0: req += f" {choices} questions with 'internal_choice_question_text'." if qt.get('id') == 'mcq' or 'choice' in label.lower(): req += " Include 4 options." pattern_str += req + "\n" total_requested_questions += count num_sections += 1 diff_label = "Easy" if difficulty_score < 30 else "Hard" if difficulty_score > 70 else "Medium" adv_str = "" if advanced.get('answerKey'): adv_str += "- Provide correct_answer.\n" if advanced.get('markingScheme'): adv_str += "- Provide marking_scheme.\n" if advanced.get('timeEstimate'): adv_str += "- Provide estimated_time_mins.\n" user_prompt = f""" Analyze content and generate exam JSON. CONTENT: {content[:15000]} BLOOMS: {blooms_str} DIFFICULTY: {diff_label} SECTIONS: {pattern_str} {adv_str} JSON STRUCTURE: {{ title: str, sections: [{{ section_name: str, instructions: str, questions: [{{ question_text: str, ... }}] }}] }} """ for attempt in range(retry_count + 1): try: completion = client.chat.completions.create( model=model_name, messages=[{"role": "system", "content": system_prompt}, {"role": "user", "content": user_prompt}], temperature=0.7, response_format={"type": "json_object"} ) raw_result = json.loads(completion.choices[0].message.content) # Use Pydantic to validate and clean the data validated_paper = ExamPaper(**raw_result) return validated_paper.model_dump() except ValidationError as ve: print(f"Pydantic Validation Error (Attempt {attempt+1}): {ve}") if attempt == retry_count: return {"error": "AI generated invalid data structure after multiple attempts."} except Exception as e: if "rate_limit" in str(e).lower(): time.sleep(2) # Short wait before retry continue return {"error": str(e)} return {"error": "Failed to generate valid exam paper."} def regenerate_single_question(content, question_data, context_paper_title=""): """ Regenerates a single specific question with pydantic validation. """ if not API_KEY or not client: return {"error": "Groq API Key not configured"} model_name = "llama-3.3-70b-versatile" system_prompt = "You are an expert academic examiner. Output ONLY raw JSON question object." user_prompt = f""" Regenerate this question but keep structure. SOURCE: {content[:10000]} EXISTING: {json.dumps(question_data)} CONTEXT: {context_paper_title} """ try: completion = client.chat.completions.create( model=model_name, messages=[{"role": "system", "content": system_prompt}, {"role": "user", "content": user_prompt}], temperature=0.8, response_format={"type": "json_object"} ) raw_q = json.loads(completion.choices[0].message.content) validated_q = Question(**raw_q) return validated_q.model_dump() except Exception as e: return {"error": str(e)}