diff --git a/core/ai_engine.py b/core/ai_engine.py
index 1e1af823..c9c30342 100644
--- a/core/ai_engine.py
+++ b/core/ai_engine.py
@@ -5,129 +5,286 @@
 import google.generativeai as genai
 from meta_ai_api import MetaAI
 
+# Configure logging
 logging.basicConfig(
     filename="karbon_ai_errors.log",
     level=logging.INFO,
     format="%(asctime)s %(levelname)s %(message)s"
 )
 
+# Global AI status
 ai_status = {"state": "connecting", "message": "Connecting to AI service..."}
 
-
 def set_ai_status(state: str, message: str):
+    """Update AI status with proper error handling."""
     ai_status["state"] = state
     ai_status["message"] = message
     logging.info(f"AI Status Updated: {state} - {message}")
 
-
 def extract_json(response: str) -> dict:
-    clean_response = response.strip()
-    if clean_response.startswith("```json"):
-        clean_response = clean_response[len("```json"):].strip()
-    if clean_response.endswith("```"):
-        clean_response = clean_response[:-len("```")].strip()
+    """
+    Extract and parse JSON from the AI response with improved error handling.
+    """
+    if not response or response.isspace():
+        logging.error("Empty AI response received")
+        raise ValueError("Empty response from AI service")
+    
+    logging.info("Raw AI response length: %d", len(response))
+    
+    # Clean the response more thoroughly
+    cleaned_response = response.strip()
+    
+    # Remove markdown code blocks
+    cleaned_response = re.sub(r'```json\s*', '', cleaned_response, flags=re.IGNORECASE)
+    cleaned_response = re.sub(r'```\s*$', '', cleaned_response, flags=re.MULTILINE)
+    cleaned_response = re.sub(r'^```|```$', '', cleaned_response, flags=re.MULTILINE)
+    
+    # Remove any leading/trailing non-JSON text
+    cleaned_response = cleaned_response.strip()
+    
+    logging.info("Cleaned response preview: %s", cleaned_response[:200] + "..." if len(cleaned_response) > 200 else cleaned_response)
+    
+    # Try multiple JSON extraction strategies
+    strategies = [
+        lambda x: x,  # Try as-is first
+        lambda x: find_json_block(x),  # Find JSON block
+        lambda x: extract_between_braces(x),  # Extract content between first { and last }
+    ]
+    
+    for i, strategy in enumerate(strategies):
+        try:
+            candidate = strategy(cleaned_response)
+            if not candidate:
+                continue
+                
+            logging.info(f"Strategy {i+1} candidate: %s", candidate[:200] + "..." if len(candidate) > 200 else candidate)
+            
+            # Parse JSON
+            parsed = json.loads(candidate)
+            
+            # Validate structure
+            if validate_json_structure(parsed):
+                logging.info("Successfully extracted and validated JSON")
+                return parsed
+            else:
+                logging.warning(f"Strategy {i+1}: JSON structure validation failed")
+                
+        except json.JSONDecodeError as e:
+            logging.warning(f"Strategy {i+1}: JSON decode failed - {e}")
+            continue
+        except Exception as e:
+            logging.warning(f"Strategy {i+1}: General error - {e}")
+            continue
+    
+    # If all strategies fail, try to create a minimal valid response
+    logging.error("All JSON extraction strategies failed")
+    raise ValueError("Failed to extract valid JSON from AI response")
 
-    try:
-        return json.loads(clean_response)
-    except json.JSONDecodeError as e:
-        logging.error(f"JSONDecodeError on cleaned string: {e}")
-        match = re.search(r'\{.*\}', clean_response, re.DOTALL)
-        if match:
-            try:
-                return json.loads(match.group(0))
-            except json.JSONDecodeError as e_inner:
-                logging.error(f"JSONDecodeError during regex fallback: {e_inner}")
-        logging.error(f"No valid JSON found in response: {response}")
-        return None
+def find_json_block(text: str) -> str:
+    """Find the first complete JSON object in the text."""
+    brace_count = 0
+    start_idx = None
+    
+    for i, char in enumerate(text):
+        if char == '{':
+            if brace_count == 0:
+                start_idx = i
+            brace_count += 1
+        elif char == '}':
+            brace_count -= 1
+            if brace_count == 0 and start_idx is not None:
+                return text[start_idx:i + 1]
+    
+    return ""
 
+def extract_between_braces(text: str) -> str:
+    """Extract content between the first { and last }."""
+    first_brace = text.find('{')
+    last_brace = text.rfind('}')
+    
+    if first_brace != -1 and last_brace != -1 and first_brace < last_brace:
+        return text[first_brace:last_brace + 1]
+    
+    return ""
 
-def generate_code_from_prompt(prompt: str, api_key: str = None, retries=2) -> str:
-    formatted = (
-        f"You are a helpful assistant that writes complete frontend apps.\n"
-        f"Given the task: \"{prompt}\"\n"
-        f"Respond ONLY in this JSON format, with no additional text, markdown, or explanation before or after the JSON:\n"
+def validate_json_structure(parsed: dict) -> bool:
+    """Validate that the parsed JSON has the expected structure."""
+    if not isinstance(parsed, dict):
+        return False
+    
+    required_keys = {"html", "css", "js", "name"}
+    if not required_keys.issubset(parsed.keys()):
+        logging.warning(f"Missing required keys. Expected: {required_keys}, Got: {set(parsed.keys())}")
+        return False
+    
+    # Check that values are strings
+    for key in required_keys:
+        if not isinstance(parsed[key], str):
+            logging.warning(f"Key '{key}' is not a string: {type(parsed[key])}")
+            return False
+    
+    return True
+
+def create_fallback_response(prompt: str) -> dict:
+    """Create a basic fallback response when AI fails."""
+    return {
+        "html": f"<!DOCTYPE html><html><head><title>Generated App</title></head><body><h1>Welcome</h1><p>App generated from: {prompt[:100]}{'...' if len(prompt) > 100 else ''}</p></body></html>",
+        "css": "body { font-family: Arial, sans-serif; margin: 40px; background: #f5f5f5; }",
+        "js": "console.log('App loaded successfully');",
+        "name": "Generated App"
+    }
+
+def generate_code_from_prompt(prompt: str, api_key: str = None, retries: int = 3) -> str:
+    """
+    Generate code from prompt with improved error handling and fallbacks.
+    """
+    formatted_prompt = (
+        f"Create a complete web application for: {prompt}\n\n"
+        f"Respond with ONLY a JSON object in this exact format (no markdown, no extra text):\n"
         "{\n"
-        "  \"html\": \"<html>...</html>\",\n"
-        "  \"css\": \"body { ... }\",\n"
-        "  \"js\": \"document.addEventListener(...)\",\n"
-        "  \"name\": \"App Name\"\n"
-        "}"
+        '  "html": "<!DOCTYPE html><html>...</html>",\n'
+        '  "css": "body { ... }",\n'
+        '  "js": "// JavaScript code here",\n'
+        '  "name": "App Name"\n'
+        "}\n\n"
+        "CRITICAL: Return ONLY the JSON object, no explanations, no markdown formatting."
     )
-
+    
+    last_error = None
+    
     for attempt in range(retries + 1):
         try:
-            set_ai_status("generating", "Generating code...")
+            set_ai_status("generating", f"Generating code (attempt {attempt + 1})...")
+            response = None
+            
+            # Try Gemini first if API key provided
             if api_key:
-                # Try Gemini
                 try:
                     genai.configure(api_key=api_key)
-                    model = genai.GenerativeModel('gemini-2.5-flash')
-                    response = model.generate_content(formatted).text
-                    logging.info(f"[Gemini] Raw AI response: {response}")
+                    model = genai.GenerativeModel('gemini-2.0-flash-exp')
+                    result = model.generate_content(formatted_prompt)
+                    response = result.text
+                    logging.info("Gemini response received")
                 except Exception as gem_e:
-                    logging.warning(f"[Gemini Fallback] Gemini failed: {gem_e}. Falling back to Meta AI.")
-                    api_key = None  # Force fallback
-                    continue
-            if not api_key:
-                # Fallback to Meta AI
-                ai = MetaAI()
-                result = ai.prompt(message=formatted)
-                response = result.get("message", "")
-                logging.info(f"[MetaAI] Raw AI response: {response}")
-
+                    logging.warning(f"Gemini failed: {gem_e}")
+                    last_error = gem_e
+            
+            # Fallback to Meta AI
+            if not response:
+                try:
+                    ai = MetaAI()
+                    result = ai.prompt(message=formatted_prompt)
+                    response = result.get("message", "")
+                    logging.info("Meta AI response received")
+                except Exception as meta_e:
+                    logging.warning(f"Meta AI failed: {meta_e}")
+                    last_error = meta_e
+            
+            if not response:
+                raise Exception("No response from any AI service")
+            
+            # Extract and validate JSON
             parsed = extract_json(response)
-            if not parsed:
-                raise ValueError("AI response couldn't be parsed into JSON.")
-
-            set_ai_status("online", "AI service is online.")
-            html = str(parsed.get("html", ""))
-            css = str(parsed.get("css", ""))
-            js = str(parsed.get("js", ""))
-
-            final_code = html.replace("</head>", f"<style>{css}</style></head>") \
-                             .replace("</body>", f"<script>{js}</script></body>")
-            logging.info(f"Final inlined HTML code (first 500 chars): {str(final_code)[:500]}...")
-            return final_code
+            
+            # Create final HTML with inlined CSS and JS
+            html = parsed.get("html", "")
+            css = parsed.get("css", "")
+            js = parsed.get("js", "")
+            
+            # Ensure HTML has proper structure
+            if not html.strip().startswith("<!DOCTYPE html>"):
+                html = f"<!DOCTYPE html><html><head><title>{parsed.get('name', 'Generated App')}</title></head><body>{html}</body></html>"
+            
+            # Inline CSS and JS
+            if css and "<style>" not in html:
+                html = html.replace("</head>", f"<style>{css}</style></head>")
+            
+            if js and "<script>" not in html:
+                html = html.replace("</body>", f"<script>{js}</script></body>")
+            
+            set_ai_status("online", "AI service is online")
+            logging.info("Successfully generated code")
+            return html
+            
         except Exception as e:
-            logging.error(f"[AI Error] Attempt {attempt + 1} failed: {str(e)}")
-            set_ai_status("error", f"AI error: {str(e)}")
-            import time
-            time.sleep(2 ** attempt)
-
-    set_ai_status("offline", "All attempts to use AI failed.")
-    return "<!DOCTYPE html><html><head><title>Error</title><style></style></head><body><h1>AI service is currently unavailable.</h1></body></html>"
-
+            logging.error(f"Attempt {attempt + 1} failed: {str(e)}")
+            last_error = e
+            set_ai_status("error", f"Attempt {attempt + 1} failed: {str(e)}")
+            
+            # Wait before retry (exponential backoff)
+            if attempt < retries:
+                import time
+                wait_time = min(2 ** attempt, 10)  # Cap at 10 seconds
+                time.sleep(wait_time)
+    
+    # All attempts failed - create fallback response
+    logging.error("All attempts failed, creating fallback response")
+    set_ai_status("offline", f"AI service unavailable: {str(last_error)}")
+    
+    fallback = create_fallback_response(prompt)
+    return fallback["html"].replace("</head>", f"<style>{fallback['css']}</style></head>").replace("</body>", f"<script>{fallback['js']}</script></body>")
 
 def optimize_prompt(prompt: str, api_key: str = None) -> str:
-    print("[optimize_prompt] Called with:", prompt)
+    """
+    Optimize vague prompts with better error handling.
+    """
+    print(f"[optimize_prompt] Input: {prompt}")
+    
+    # If prompt is already detailed enough, return as-is
     if len(prompt.strip()) >= 20 and not is_generic(prompt):
-        return prompt
-
+        return prompt.strip()
+    
+    optimization_prompt = (
+        f"Transform this brief UI request into a detailed web development specification: '{prompt.strip()}'\n\n"
+        f"Include:\n"
+        f"- Specific layout and components\n"
+        f"- Visual styling preferences\n"
+        f"- Any interactive features\n"
+        f"- Target audience considerations\n\n"
+        f"Return only the enhanced prompt, no explanations."
+    )
+    
     try:
         if api_key:
             genai.configure(api_key=api_key)
-            model = genai.GenerativeModel('gemini-2.5-flash')
-            enriched = model.generate_content(
-                f"Transform the following vague or minimal UI prompt into a clear, detailed, and professional instruction specifically for frontend web development. "
-                f"Focus on HTML, CSS, and JavaScript. Include layout details, components to be included, styling considerations, and any interactivity if relevant. "
-                f"Keep the improved prompt concise yet specific. Do not add commentary or formatting:\n\n"
-                f"User Prompt: \"{prompt.strip()}\"\n\n"
-                f"Refined Prompt:"
-            ).text.strip()
-            if enriched:
-                return enriched
+            model = genai.GenerativeModel('gemini-2.0-flash-exp')
+            result = model.generate_content(optimization_prompt)
+            enhanced = result.text.strip()
+            
+            if enhanced and len(enhanced) > len(prompt):
+                logging.info("Prompt successfully optimized with Gemini")
+                return enhanced
+                
     except Exception as e:
-        logging.error(f"optimize_prompt failed with Gemini: {str(e)}")
+        logging.warning(f"Prompt optimization failed: {e}")
+    
+    # Fallback to rule-based enhancement
     return rule_based_enhancement(prompt)
 
-
 def rule_based_enhancement(prompt: str) -> str:
+    """
+    Rule-based prompt enhancement as fallback.
+    """
     prompt = prompt.strip()
-    return f"{prompt}\n\nUse semantic HTML5 structure.\nApply responsive CSS (mobile-first).\nInclude clean modular JavaScript with comments."
-
+    
+    enhancements = [
+        "Use modern HTML5 semantic structure with proper accessibility.",
+        "Apply responsive CSS with mobile-first approach and modern design principles.",
+        "Include interactive JavaScript features with smooth animations.",
+        "Ensure clean, professional styling with good typography and color scheme."
+    ]
+    
+    return f"{prompt}\n\n" + " ".join(enhancements)
 
 def is_generic(prompt: str) -> bool:
+    """
+    Check if prompt is too generic and needs enhancement.
+    """
     generic_phrases = {
-        "make a website", "build ui", "create page", "webpage", "dashboard", "login", "landing page"
+        "make a website", "build ui", "create page", "webpage", 
+        "dashboard", "login", "landing page", "app", "site"
     }
-    return prompt.lower().strip() in generic_phrases
+    
+    prompt_lower = prompt.lower().strip()
+    return any(phrase in prompt_lower for phrase in generic_phrases) and len(prompt_lower) < 30
+
diff --git a/requirements.txt b/requirements.txt
index 383c51d6..bbf45ae6 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -5,3 +5,4 @@ PyGithub
 GitPython
 meta-ai-api
 tkhtmlview
+requests>=2.31.0