Update app.py

app.py

@@ -1,13 +1,519 @@
+# import pickle
+# import faiss
+# from flask import Flask, request, jsonify, render_template_string
+# from sentence_transformers import SentenceTransformer
+# from transformers import AutoTokenizer, AutoModelForCausalLM
+# from huggingface_hub import hf_hub_download
+# import torch
+# import os
+# from functools import lru_cache
+# import hashlib
+
+# app = Flask(__name__)
+
+# print("=" * 50)
+# print("Loading models and data...")
+# print("=" * 50)
+
+# # ------------------------------
+# # Load embedding model (CPU)
+# # ------------------------------
+# embed_model = SentenceTransformer("all-MiniLM-L6-v2")
+# print("✓ Embedding model loaded")
+
+# # ------------------------------
+# # Download files from Hugging Face
+# # ------------------------------
+# REPO_ID = "Redfire-1234/pcb_tutor"
+
+# print("Downloading subject files from Hugging Face...")
+
+# # Download Biology files
+# bio_chunks_path = hf_hub_download(repo_id=REPO_ID, filename="bio_chunks.pkl", repo_type="model")
+# faiss_bio_path = hf_hub_download(repo_id=REPO_ID, filename="faiss_bio.bin", repo_type="model")
+
+# # Download Chemistry files
+# chem_chunks_path = hf_hub_download(repo_id=REPO_ID, filename="chem_chunks.pkl", repo_type="model")
+# faiss_chem_path = hf_hub_download(repo_id=REPO_ID, filename="faiss_chem.bin", repo_type="model")
+
+# # Download Physics files
+# phy_chunks_path = hf_hub_download(repo_id=REPO_ID, filename="phy_chunks.pkl", repo_type="model")
+# faiss_phy_path = hf_hub_download(repo_id=REPO_ID, filename="faiss_phy.bin", repo_type="model")
+
+# # Load all subjects into memory
+# SUBJECTS = {
+#     "biology": {
+#         "chunks": pickle.load(open(bio_chunks_path, "rb")),
+#         "index": faiss.read_index(faiss_bio_path)
+#     },
+#     "chemistry": {
+#         "chunks": pickle.load(open(chem_chunks_path, "rb")),
+#         "index": faiss.read_index(faiss_chem_path)
+#     },
+#     "physics": {
+#         "chunks": pickle.load(open(phy_chunks_path, "rb")),
+#         "index": faiss.read_index(faiss_phy_path)
+#     }
+# }
+
+# print(f"✓ Biology: {len(SUBJECTS['biology']['chunks'])} chunks loaded")
+# print(f"✓ Chemistry: {len(SUBJECTS['chemistry']['chunks'])} chunks loaded")
+# print(f"✓ Physics: {len(SUBJECTS['physics']['chunks'])} chunks loaded")
+
+# # ------------------------------
+# # Load LLM model (CPU) with optimizations
+# # ------------------------------
+# model_name = "Qwen/Qwen2.5-3B-Instruct"
+# print(f"Loading LLM: {model_name}")
+# tokenizer = AutoTokenizer.from_pretrained(model_name)
+# device = "cpu"
+
+# # OPTIMIZATION: Load model with better dtype for CPU
+# model = AutoModelForCausalLM.from_pretrained(
+#     model_name,
+#     torch_dtype=torch.float32,
+#     low_cpu_mem_usage=True  # Optimization: Better memory management
+# ).to(device)
+
+# # OPTIMIZATION: Set model to eval mode and optimize for inference
+# model.eval()
+# if hasattr(torch, 'set_num_threads'):
+#     torch.set_num_threads(4)  # Optimization: Use multiple CPU threads
+
+# print(f"✓ LLM loaded on {device}")
+
+# print("=" * 50)
+# print("All models loaded successfully!")
+# print("=" * 50)
+
+# # ------------------------------
+# # OPTIMIZATION: Add caching for MCQ generation
+# # ------------------------------
+# MCQ_CACHE = {}
+# MAX_CACHE_SIZE = 100
+
+# def get_cache_key(topic, subject, context_hash):
+#     """Generate a unique cache key"""
+#     return f"{subject}:{topic}:{context_hash}"
+
+# def cache_mcq(key, mcqs):
+#     """Cache generated MCQs with size limit"""
+#     if len(MCQ_CACHE) >= MAX_CACHE_SIZE:
+#         # Remove oldest entry
+#         MCQ_CACHE.pop(next(iter(MCQ_CACHE)))
+#     MCQ_CACHE[key] = mcqs
+
+# # ------------------------------
+# # RAG Search in specific subject (optimized)
+# # ------------------------------
+# def rag_search(query, subject, k=5):
+#     if subject not in SUBJECTS:
+#         return None
+    
+#     chunks = SUBJECTS[subject]["chunks"]
+#     index = SUBJECTS[subject]["index"]
+    
+#     # OPTIMIZATION: Encode query (already fast with sentence-transformers)
+#     q_emb = embed_model.encode([query], show_progress_bar=False).astype("float32")
+#     D, I = index.search(q_emb, k)
+    
+#     # Get the actual chunks
+#     results = []
+#     for idx in I[0]:
+#         if idx < len(chunks):
+#             results.append(chunks[idx])
+    
+#     return "\n\n".join(results)
+
+# # ------------------------------
+# # OPTIMIZED MCQ Generation with reduced tokens
+# # ------------------------------
+# def generate_mcqs(context, topic, subject):
+#     # OPTIMIZATION: Check cache first
+#     context_hash = hashlib.md5(context.encode()).hexdigest()[:8]
+#     cache_key = get_cache_key(topic, subject, context_hash)
+    
+#     if cache_key in MCQ_CACHE:
+#         print("✓ Using cached MCQs")
+#         return MCQ_CACHE[cache_key]
+    
+#     # OPTIMIZATION: Shortened prompt for faster generation
+#     prompt = f"""You are a Class-12 {subject.title()} teacher creating MCQs.
+# Topic: "{topic}"
+# Context:
+# {context}
+
+# Generate exactly 5 MCQs in this format:
+# Q1. [Question]
+# A) [Option]
+# B) [Option]
+# C) [Option]
+# D) [Option]
+# Correct Answer: [Letter] - [Reason]
+
+# Rules: Make correct answer from context, realistic distractors.
+# Generate 5 MCQs:"""
+    
+#     # OPTIMIZATION: Reduced max_length for faster tokenization
+#     inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=1536).to(device)
+    
+#     # OPTIMIZATION: Use torch.no_grad() for inference (saves memory)
+#     with torch.no_grad():
+#         # OPTIMIZATION: Reduced max_new_tokens from 900 to 600 (sufficient for 5 MCQs)
+#         # OPTIMIZATION: Reduced temperature from 0.15 to 0.1 (faster, more deterministic)
+#         # OPTIMIZATION: Added num_beams=1 (greedy decoding, faster than sampling)
+#         outputs = model.generate(
+#             **inputs,
+#             max_new_tokens=600,  # Reduced from 900
+#             temperature=0.1,     # Reduced from 0.15
+#             top_p=0.85,          # Slightly adjusted
+#             do_sample=True,
+#             repetition_penalty=1.15,
+#             pad_token_id=tokenizer.eos_token_id  # Optimization: Explicit pad token
+#         )
+    
+#     result = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    
+#     # Extract only the generated MCQs
+#     if "Generate 5 MCQs:" in result:
+#         result = result.split("Generate 5 MCQs:")[-1].strip()
+    
+#     # OPTIMIZATION: Cache the result
+#     cache_mcq(cache_key, result)
+    
+#     return result
+
+# def verify_and_correct_answers(mcqs_text, context):
+#     """
+#     This function is kept for future enhancements
+#     """
+#     return mcqs_text
+
+# # ------------------------------
+# # HTML UI (with improved loading message)
+# # ------------------------------
+# HTML_TEMPLATE = """
+# <!DOCTYPE html>
+# <html>
+# <head>
+#     <title>Class 12 PCB MCQ Generator</title>
+#     <style>
+#         * { margin: 0; padding: 0; box-sizing: border-box; }
+#         body {
+#             font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif;
+#             background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
+#             min-height: 100vh;
+#             padding: 20px;
+#         }
+#         .container {
+#             max-width: 900px;
+#             margin: 0 auto;
+#             background: white;
+#             border-radius: 20px;
+#             box-shadow: 0 20px 60px rgba(0,0,0,0.3);
+#             overflow: hidden;
+#         }
+#         .header {
+#             background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
+#             color: white;
+#             padding: 30px;
+#             text-align: center;
+#         }
+#         .header h1 { font-size: 2.5em; margin-bottom: 10px; }
+#         .content { padding: 40px; }
+#         .form-group {
+#             margin-bottom: 25px;
+#         }
+#         label {
+#             display: block;
+#             font-weight: 600;
+#             margin-bottom: 10px;
+#             color: #333;
+#             font-size: 16px;
+#         }
+#         select, input {
+#             width: 100%;
+#             padding: 15px;
+#             border: 2px solid #e0e0e0;
+#             border-radius: 10px;
+#             font-size: 16px;
+#             font-family: inherit;
+#             transition: border-color 0.3s;
+#         }
+#         select:focus, input:focus {
+#             outline: none;
+#             border-color: #667eea;
+#         }
+#         button {
+#             width: 100%;
+#             padding: 18px;
+#             background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
+#             color: white;
+#             border: none;
+#             border-radius: 10px;
+#             font-size: 18px;
+#             font-weight: 600;
+#             cursor: pointer;
+#             transition: all 0.3s;
+#         }
+#         button:hover {
+#             transform: translateY(-2px);
+#             box-shadow: 0 10px 20px rgba(102, 126, 234, 0.4);
+#         }
+#         button:disabled {
+#             background: #ccc;
+#             cursor: not-allowed;
+#             transform: none;
+#         }
+#         .result {
+#             margin-top: 30px;
+#             padding: 25px;
+#             background: #f8f9fa;
+#             border-radius: 10px;
+#             border-left: 4px solid #667eea;
+#             display: none;
+#         }
+#         .result.show { display: block; }
+#         .result h3 {
+#             color: #667eea;
+#             margin-bottom: 20px;
+#             font-size: 1.4em;
+#         }
+#         .mcq-content {
+#             background: white;
+#             padding: 25px;
+#             border-radius: 8px;
+#             white-space: pre-wrap;
+#             line-height: 1.9;
+#             font-size: 15px;
+#         }
+#         .loading {
+#             text-align: center;
+#             padding: 30px;
+#             display: none;
+#         }
+#         .loading.show { display: block; }
+#         .spinner {
+#             border: 4px solid #f3f3f3;
+#             border-top: 4px solid #667eea;
+#             border-radius: 50%;
+#             width: 50px;
+#             height: 50px;
+#             animation: spin 1s linear infinite;
+#             margin: 0 auto 15px;
+#         }
+#         @keyframes spin {
+#             0% { transform: rotate(0deg); }
+#             100% { transform: rotate(360deg); }
+#         }
+#         .subject-tag {
+#             display: inline-block;
+#             padding: 5px 15px;
+#             border-radius: 20px;
+#             font-size: 13px;
+#             font-weight: 600;
+#             margin-right: 10px;
+#         }
+#         .bio { background: #d4edda; color: #155724; }
+#         .chem { background: #d1ecf1; color: #0c5460; }
+#         .phy { background: #f8d7da; color: #721c24; }
+#         .optimization-badge {
+#             background: #28a745;
+#             color: white;
+#             padding: 5px 12px;
+#             border-radius: 15px;
+#             font-size: 12px;
+#             margin-left: 10px;
+#         }
+#     </style>
+# </head>
+# <body>
+#     <div class="container">
+#         <div class="header">
+#             <h1>🎓 Class 12 PCB MCQ Generator</h1>
+#             <p style="font-size: 1.1em; margin-bottom: 15px;">Generate practice MCQs from your textbooks <span class="optimization-badge">⚡ Optimized</span></p>
+#             <div>
+#                 <span class="subject-tag bio">Biology</span>
+#                 <span class="subject-tag chem">Chemistry</span>
+#                 <span class="subject-tag phy">Physics</span>
+#             </div>
+#         </div>
+        
+#         <div class="content">
+#             <div class="form-group">
+#                 <label for="subject">📚 Select Subject</label>
+#                 <select id="subject">
+#                     <option value="biology">Biology</option>
+#                     <option value="chemistry">Chemistry</option>
+#                     <option value="physics">Physics</option>
+#                 </select>
+#             </div>
+            
+#             <div class="form-group">
+#                 <label for="topic">✏️ Enter Topic</label>
+#                 <input type="text" id="topic" placeholder="e.g., Mitochondria, Chemical Bonding, Newton's Laws">
+#             </div>
+            
+#             <button onclick="generateMCQs()">🚀 Generate 5 MCQs</button>
+            
+#             <div class="loading" id="loading">
+#                 <div class="spinner"></div>
+#                 <p style="color: #666; font-size: 16px;">Generating MCQs... This may take 20-40 seconds</p>
+#                 <p style="color: #999; font-size: 13px; margin-top: 10px;">⚡ Optimized for faster generation</p>
+#             </div>
+            
+#             <div class="result" id="result">
+#                 <h3>📝 Generated MCQs:</h3>
+#                 <div style="background: #fff3cd; padding: 12px; border-radius: 6px; margin-bottom: 15px; color: #856404; font-size: 14px;">
+#                     ⚠️ <strong>Note:</strong> AI-generated answers may occasionally be incorrect. Please verify answers using your textbook.
+#                 </div>
+#                 <div class="mcq-content" id="mcqContent"></div>
+#             </div>
+#         </div>
+#     </div>
+#     <script>
+#         async function generateMCQs() {
+#             const subject = document.getElementById('subject').value;
+#             const topic = document.getElementById('topic').value.trim();
+            
+#             if (!topic) {
+#                 alert('⚠️ Please enter a topic!');
+#                 return;
+#             }
+            
+#             const loading = document.getElementById('loading');
+#             const result = document.getElementById('result');
+#             const btn = document.querySelector('button');
+            
+#             loading.classList.add('show');
+#             result.classList.remove('show');
+#             btn.disabled = true;
+#             btn.textContent = '⏳ Generating...';
+            
+#             try {
+#                 const response = await fetch('/generate', {
+#                     method: 'POST',
+#                     headers: {'Content-Type': 'application/json'},
+#                     body: JSON.stringify({subject, topic})
+#                 });
+                
+#                 const data = await response.json();
+                
+#                 if (data.error) {
+#                     alert('❌ Error: ' + data.error);
+#                     return;
+#                 }
+                
+#                 document.getElementById('mcqContent').textContent = data.mcqs;
+#                 result.classList.add('show');
+#             } catch (error) {
+#                 alert('❌ Error: ' + error.message);
+#             } finally {
+#                 loading.classList.remove('show');
+#                 btn.disabled = false;
+#                 btn.textContent = '🚀 Generate 5 MCQs';
+#             }
+#         }
+        
+#         // Allow Enter key to submit
+#         document.getElementById('topic').addEventListener('keypress', function(e) {
+#             if (e.key === 'Enter') {
+#                 generateMCQs();
+#             }
+#         });
+#     </script>
+# </body>
+# </html>
+# """
+
+# # ------------------------------
+# # Routes
+# # ------------------------------
+# @app.route("/")
+# def home():
+#     return render_template_string(HTML_TEMPLATE)
+
+# @app.route("/generate", methods=["POST"])
+# def generate():
+#     try:
+#         data = request.json
+#         subject = data.get("subject", "").lower()
+#         topic = data.get("topic", "")
+        
+#         if not topic:
+#             return jsonify({"error": "Topic is required"}), 400
+        
+#         if subject not in SUBJECTS:
+#             return jsonify({"error": "Invalid subject. Choose biology, chemistry, or physics."}), 400
+        
+#         print(f"\n🔍 Searching {subject} for topic: {topic}")
+        
+#         # Retrieve context from RAG
+#         context = rag_search(topic, subject, k=5)
+        
+#         if not context or len(context.strip()) < 50:
+#             return jsonify({"error": f"No relevant content found in {subject} for topic: {topic}"}), 404
+        
+#         print(f"✓ Found context ({len(context)} chars)")
+        
+#         # Generate MCQs (now with caching)
+#         print("🤖 Generating MCQs...")
+#         mcqs = generate_mcqs(context, topic, subject)
+        
+#         print("✓ MCQs generated successfully")
+        
+#         return jsonify({"mcqs": mcqs, "subject": subject})
+    
+#     except Exception as e:
+#         print(f"❌ Error in /generate: {e}")
+#         import traceback
+#         traceback.print_exc()
+#         return jsonify({"error": str(e)}), 500
+
+# @app.route("/health")
+# def health():
+#     return jsonify({
+#         "status": "healthy",
+#         "subjects": {
+#             "biology": len(SUBJECTS["biology"]["chunks"]),
+#             "chemistry": len(SUBJECTS["chemistry"]["chunks"]),
+#             "physics": len(SUBJECTS["physics"]["chunks"])
+#         },
+#         "cache_size": len(MCQ_CACHE)
+#     })
+
+# # OPTIMIZATION: Add cache stats endpoint
+# @app.route("/cache/stats")
+# def cache_stats():
+#     return jsonify({
+#         "cached_topics": len(MCQ_CACHE),
+#         "max_cache_size": MAX_CACHE_SIZE,
+#         "cache_keys": list(MCQ_CACHE.keys())
+#     })
+
+# # OPTIMIZATION: Add cache clear endpoint (optional)
+# @app.route("/cache/clear", methods=["POST"])
+# def clear_cache():
+#     MCQ_CACHE.clear()
+#     return jsonify({"status": "Cache cleared"})
+
+# # ------------------------------
+# # Run the App
+# # ------------------------------
+# if __name__ == "__main__":
+#     port = int(os.environ.get("PORT", 7860))
+#     print(f"\n🚀 Starting Flask on 0.0.0.0:{port}\n")
+#     app.run(host="0.0.0.0", port=port, debug=False)
+
 import pickle
 import faiss
 from flask import Flask, request, jsonify, render_template_string
 from sentence_transformers import SentenceTransformer
-from transformers import AutoTokenizer, AutoModelForCausalLM
 from huggingface_hub import hf_hub_download
-import torch
-import os
-from functools import lru_cache
 import hashlib
+import re
+import os
+from groq import Groq
 
 app = Flask(__name__)
 
@@ -15,6 +521,18 @@ print("=" * 50)
 print("Loading models and data...")
 print("=" * 50)
 
+# ------------------------------
+# Initialize Groq API Client
+# ------------------------------
+# Get your free API key from: https://console.groq.com/keys
+GROQ_API_KEY = os.environ.get("GROQ_API_KEY", "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")  # Set this in your environment
+if not GROQ_API_KEY:
+    print("⚠️ WARNING: GROQ_API_KEY not set. Get one from https://console.groq.com/keys")
+    print("Set it with: export GROQ_API_KEY='your-key-here'")
+
+groq_client = Groq(api_key=GROQ_API_KEY)
+print("✓ Groq API client initialized")
+
 # ------------------------------
 # Load embedding model (CPU)
 # ------------------------------
@@ -60,34 +578,12 @@ print(f"✓ Biology: {len(SUBJECTS['biology']['chunks'])} chunks loaded")
 print(f"✓ Chemistry: {len(SUBJECTS['chemistry']['chunks'])} chunks loaded")
 print(f"✓ Physics: {len(SUBJECTS['physics']['chunks'])} chunks loaded")
 
-# ------------------------------
-# Load LLM model (CPU) with optimizations
-# ------------------------------
-model_name = "Qwen/Qwen2.5-3B-Instruct"
-print(f"Loading LLM: {model_name}")
-tokenizer = AutoTokenizer.from_pretrained(model_name)
-device = "cpu"
-
-# OPTIMIZATION: Load model with better dtype for CPU
-model = AutoModelForCausalLM.from_pretrained(
-    model_name,
-    torch_dtype=torch.float32,
-    low_cpu_mem_usage=True  # Optimization: Better memory management
-).to(device)
-
-# OPTIMIZATION: Set model to eval mode and optimize for inference
-model.eval()
-if hasattr(torch, 'set_num_threads'):
-    torch.set_num_threads(4)  # Optimization: Use multiple CPU threads
-
-print(f"✓ LLM loaded on {device}")
-
 print("=" * 50)
 print("All models loaded successfully!")
 print("=" * 50)
 
 # ------------------------------
-# OPTIMIZATION: Add caching for MCQ generation
+# Caching for MCQ generation
 # ------------------------------
 MCQ_CACHE = {}
 MAX_CACHE_SIZE = 100
@@ -99,12 +595,11 @@ def get_cache_key(topic, subject, context_hash):
 def cache_mcq(key, mcqs):
     """Cache generated MCQs with size limit"""
     if len(MCQ_CACHE) >= MAX_CACHE_SIZE:
-        # Remove oldest entry
         MCQ_CACHE.pop(next(iter(MCQ_CACHE)))
     MCQ_CACHE[key] = mcqs
 
 # ------------------------------
-# RAG Search in specific subject (optimized)
+# RAG Search in specific subject
 # ------------------------------
 def rag_search(query, subject, k=5):
     if subject not in SUBJECTS:
@@ -113,11 +608,9 @@ def rag_search(query, subject, k=5):
     chunks = SUBJECTS[subject]["chunks"]
     index = SUBJECTS[subject]["index"]
     
-    # OPTIMIZATION: Encode query (already fast with sentence-transformers)
     q_emb = embed_model.encode([query], show_progress_bar=False).astype("float32")
     D, I = index.search(q_emb, k)
     
-    # Get the actual chunks
     results = []
     for idx in I[0]:
         if idx < len(chunks):
@@ -126,10 +619,10 @@ def rag_search(query, subject, k=5):
     return "\n\n".join(results)
 
 # ------------------------------
-# OPTIMIZED MCQ Generation with reduced tokens
+# MCQ Generation using Groq API
 # ------------------------------
 def generate_mcqs(context, topic, subject):
-    # OPTIMIZATION: Check cache first
+    # Check cache first
     context_hash = hashlib.md5(context.encode()).hexdigest()[:8]
     cache_key = get_cache_key(topic, subject, context_hash)
     
@@ -137,60 +630,92 @@ def generate_mcqs(context, topic, subject):
         print("✓ Using cached MCQs")
         return MCQ_CACHE[cache_key]
     
-    # OPTIMIZATION: Shortened prompt for faster generation
-    prompt = f"""You are a Class-12 {subject.title()} teacher creating MCQs.
-Topic: "{topic}"
-Context:
-{context}
-
-Generate exactly 5 MCQs in this format:
-Q1. [Question]
-A) [Option]
-B) [Option]
-C) [Option]
-D) [Option]
-Correct Answer: [Letter] - [Reason]
-
-Rules: Make correct answer from context, realistic distractors.
-Generate 5 MCQs:"""
+    print("🤖 Generating MCQs using Groq API...")
     
-    # OPTIMIZATION: Reduced max_length for faster tokenization
-    inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=1536).to(device)
+    # Prompt for MCQ generation
+    prompt = f"""You are a Class-12 {subject.title()} teacher creating MCQs for students.
+
+Topic: "{topic}"
+
+Reference material from textbook:
+{context[:1500]}
+
+TASK: Generate exactly 5 multiple-choice questions based on the reference material above.
+
+FORMAT (follow this EXACTLY):
+Q1. [Clear question based on the material]
+A) [First option]
+B) [Second option]
+C) [Third option]
+D) [Fourth option]
+Answer: [A/B/C/D] - [Brief explanation why this is correct based on the material]
+
+REQUIREMENTS:
+- Questions must be answerable from the reference material
+- All 4 options should be plausible
+- The correct answer must be clearly supported by the material
+- Keep explanations brief (1-2 sentences)
+- Generate all 5 questions in the format above
+
+Generate 5 MCQs now:"""
     
-    # OPTIMIZATION: Use torch.no_grad() for inference (saves memory)
-    with torch.no_grad():
-        # OPTIMIZATION: Reduced max_new_tokens from 900 to 600 (sufficient for 5 MCQs)
-        # OPTIMIZATION: Reduced temperature from 0.15 to 0.1 (faster, more deterministic)
-        # OPTIMIZATION: Added num_beams=1 (greedy decoding, faster than sampling)
-        outputs = model.generate(
-            **inputs,
-            max_new_tokens=600,  # Reduced from 900
-            temperature=0.1,     # Reduced from 0.15
-            top_p=0.85,          # Slightly adjusted
-            do_sample=True,
-            repetition_penalty=1.15,
-            pad_token_id=tokenizer.eos_token_id  # Optimization: Explicit pad token
+    try:
+        # Call Groq API
+        chat_completion = groq_client.chat.completions.create(
+            messages=[
+                {
+                    "role": "system",
+                    "content": "You are an expert Class-12 teacher who creates high-quality multiple-choice questions from textbook content. You always follow the exact format specified."
+                },
+                {
+                    "role": "user",
+                    "content": prompt
+                }
+            ],
+            model="llama-3.3-70b-versatile",  # Fast and accurate
+            temperature=0.3,
+            max_tokens=1500,
+            top_p=0.9
         )
+        
+        result = chat_completion.choices[0].message.content.strip()
+        
+        # Clean the output
+        result = clean_mcq_output(result)
+        
+        # Cache the result
+        cache_mcq(cache_key, result)
+        
+        print("✓ MCQs generated successfully")
+        return result
+        
+    except Exception as e:
+        print(f"❌ Groq API Error: {e}")
+        return f"Error generating MCQs: {str(e)}\n\nPlease make sure GROQ_API_KEY is set correctly."
+
+def clean_mcq_output(text):
+    """Clean and format the MCQ output"""
+    lines = text.split('\n')
+    cleaned_lines = []
     
-    result = tokenizer.decode(outputs[0], skip_special_tokens=True)
-    
-    # Extract only the generated MCQs
-    if "Generate 5 MCQs:" in result:
-        result = result.split("Generate 5 MCQs:")[-1].strip()
-    
-    # OPTIMIZATION: Cache the result
-    cache_mcq(cache_key, result)
+    for line in lines:
+        line = line.strip()
+        
+        # Keep question lines, options, and answers
+        if (re.match(r'^Q\d+\.', line) or 
+            line.startswith(('A)', 'B)', 'C)', 'D)', 'Answer:', 'Correct Answer:')) or
+            not line):
+            
+            # Normalize answer format
+            if line.startswith('Correct Answer:'):
+                line = line.replace('Correct Answer:', 'Answer:')
+            
+            cleaned_lines.append(line)
     
-    return result
-
-def verify_and_correct_answers(mcqs_text, context):
-    """
-    This function is kept for future enhancements
-    """
-    return mcqs_text
+    return '\n'.join(cleaned_lines)
 
 # ------------------------------
-# HTML UI (with improved loading message)
+# HTML UI
 # ------------------------------
 HTML_TEMPLATE = """
 <!DOCTYPE html>
@@ -221,9 +746,7 @@ HTML_TEMPLATE = """
         }
         .header h1 { font-size: 2.5em; margin-bottom: 10px; }
         .content { padding: 40px; }
-        .form-group {
-            margin-bottom: 25px;
-        }
+        .form-group { margin-bottom: 25px; }
         label {
             display: block;
             font-weight: 600;
@@ -317,8 +840,8 @@ HTML_TEMPLATE = """
         .bio { background: #d4edda; color: #155724; }
         .chem { background: #d1ecf1; color: #0c5460; }
         .phy { background: #f8d7da; color: #721c24; }
-        .optimization-badge {
-            background: #28a745;
+        .api-badge {
+            background: #17a2b8;
             color: white;
             padding: 5px 12px;
             border-radius: 15px;
@@ -331,7 +854,10 @@ HTML_TEMPLATE = """
     <div class="container">
         <div class="header">
             <h1>🎓 Class 12 PCB MCQ Generator</h1>
-            <p style="font-size: 1.1em; margin-bottom: 15px;">Generate practice MCQs from your textbooks <span class="optimization-badge">⚡ Optimized</span></p>
+            <p style="font-size: 1.1em; margin-bottom: 15px;">
+                Generate practice MCQs from your textbooks 
+                <span class="api-badge">⚡ Powered by Llama 3.3 70B</span>
+            </p>
             <div>
                 <span class="subject-tag bio">Biology</span>
                 <span class="subject-tag chem">Chemistry</span>
@@ -358,14 +884,14 @@ HTML_TEMPLATE = """
             
             <div class="loading" id="loading">
                 <div class="spinner"></div>
-                <p style="color: #666; font-size: 16px;">Generating MCQs... This may take 20-40 seconds</p>
-                <p style="color: #999; font-size: 13px; margin-top: 10px;">⚡ Optimized for faster generation</p>
+                <p style="color: #666; font-size: 16px;">Generating high-quality MCQs...</p>
+                <p style="color: #999; font-size: 13px; margin-top: 10px;">⚡ Using Llama 3.3 70B via Groq API (5-10 seconds)</p>
             </div>
             
             <div class="result" id="result">
                 <h3>📝 Generated MCQs:</h3>
-                <div style="background: #fff3cd; padding: 12px; border-radius: 6px; margin-bottom: 15px; color: #856404; font-size: 14px;">
-                    ⚠️ <strong>Note:</strong> AI-generated answers may occasionally be incorrect. Please verify answers using your textbook.
+                <div style="background: #d4edda; padding: 12px; border-radius: 6px; margin-bottom: 15px; color: #155724; font-size: 14px;">
+                    ✓ <strong>High Accuracy:</strong> Generated by Llama 3.3 70B - One of the most capable models available!
                 </div>
                 <div class="mcq-content" id="mcqContent"></div>
             </div>
@@ -415,7 +941,6 @@ HTML_TEMPLATE = """
             }
         }
         
-        // Allow Enter key to submit
         document.getElementById('topic').addEventListener('keypress', function(e) {
             if (e.key === 'Enter') {
                 generateMCQs();
@@ -444,28 +969,23 @@ def generate():
             return jsonify({"error": "Topic is required"}), 400
         
         if subject not in SUBJECTS:
-            return jsonify({"error": "Invalid subject. Choose biology, chemistry, or physics."}), 400
+            return jsonify({"error": "Invalid subject"}), 400
         
         print(f"\n🔍 Searching {subject} for topic: {topic}")
         
-        # Retrieve context from RAG
         context = rag_search(topic, subject, k=5)
         
         if not context or len(context.strip()) < 50:
-            return jsonify({"error": f"No relevant content found in {subject} for topic: {topic}"}), 404
+            return jsonify({"error": f"No relevant content found for topic: {topic}"}), 404
         
         print(f"✓ Found context ({len(context)} chars)")
         
-        # Generate MCQs (now with caching)
-        print("🤖 Generating MCQs...")
         mcqs = generate_mcqs(context, topic, subject)
         
-        print("✓ MCQs generated successfully")
-        
         return jsonify({"mcqs": mcqs, "subject": subject})
     
     except Exception as e:
-        print(f"❌ Error in /generate: {e}")
+        print(f"❌ Error: {e}")
         import traceback
         traceback.print_exc()
         return jsonify({"error": str(e)}), 500
@@ -474,24 +994,10 @@ def generate():
 def health():
     return jsonify({
         "status": "healthy",
-        "subjects": {
-            "biology": len(SUBJECTS["biology"]["chunks"]),
-            "chemistry": len(SUBJECTS["chemistry"]["chunks"]),
-            "physics": len(SUBJECTS["physics"]["chunks"])
-        },
+        "model": "llama-3.3-70b-versatile (Groq API)",
         "cache_size": len(MCQ_CACHE)
     })
 
-# OPTIMIZATION: Add cache stats endpoint
-@app.route("/cache/stats")
-def cache_stats():
-    return jsonify({
-        "cached_topics": len(MCQ_CACHE),
-        "max_cache_size": MAX_CACHE_SIZE,
-        "cache_keys": list(MCQ_CACHE.keys())
-    })
-
-# OPTIMIZATION: Add cache clear endpoint (optional)
 @app.route("/cache/clear", methods=["POST"])
 def clear_cache():
     MCQ_CACHE.clear()
@@ -505,5 +1011,3 @@ if __name__ == "__main__":
     print(f"\n🚀 Starting Flask on 0.0.0.0:{port}\n")
     app.run(host="0.0.0.0", port=port, debug=False)
 
-
-