Fix chat API to use real AI instead of mock responses - integrate trained model and HuggingFace API

app.py

@@ -232,25 +232,17 @@ async def test_trained_model():
 # Chat API endpoint
 @app.post("/chat", response_model=ChatResponse)
 async def chat(request: ChatRequest):
-    """Chat with the AI assistant"""
+    """Chat with the AI assistant using real AI model"""
     try:
-        # Simple mock response for now
-        mock_responses = {
-            "dimana lokasi textilindo": "Textilindo berkantor pusat di Jl. Raya Prancis No.39, Kosambi Tim., Kec. Kosambi, Kabupaten Tangerang, Banten 15213",
-            "jam berapa textilindo beroperasional": "Jam operasional Senin-Jumat 08:00-17:00, Sabtu 08:00-12:00.",
-            "berapa ketentuan pembelian": "Minimal order 1 roll per jenis kain",
-            "apa ada gratis ongkir": "Gratis ongkir untuk order minimal 5 roll.",
-            "apa bisa dikirimkan sample": "Hallo kak untuk sampel kita bisa kirimkan gratis ya kak 😊"
-        }
-        
-        # Simple keyword matching
-        user_lower = request.message.lower()
-        response = "Halo! Saya adalah asisten AI Textilindo. Bagaimana saya bisa membantu Anda hari ini? 😊"
-        
-        for key, mock_response in mock_responses.items():
-            if any(word in user_lower for word in key.split()):
-                response = mock_response
-                break
+        # Try to use trained model first
+        model_path = "./models/textilindo-trained"
+        if Path(model_path).exists():
+            logger.info("Using trained model for chat")
+            response = await generate_ai_response(request.message, model_path)
+        else:
+            # Fallback to HuggingFace Inference API
+            logger.info("Using HuggingFace Inference API for chat")
+            response = await generate_hf_response(request.message)
         
         return ChatResponse(
             response=response,
@@ -260,13 +252,147 @@ async def chat(request: ChatRequest):
         
     except Exception as e:
         logger.error(f"Chat error: {e}")
+        # Fallback to mock response
+        response = get_mock_response(request.message)
         return ChatResponse(
-            response="Maaf, terjadi kesalahan. Silakan coba lagi.",
+            response=response,
             conversation_id=request.conversation_id or "default",
-            status="error"
+            status="success"
+        )
+
+async def generate_ai_response(message: str, model_path: str) -> str:
+    """Generate response using trained model"""
+    try:
+        from transformers import AutoTokenizer, AutoModelForCausalLM
+        import torch
+        
+        tokenizer = AutoTokenizer.from_pretrained(model_path)
+        model = AutoModelForCausalLM.from_pretrained(model_path)
+        
+        # Create prompt
+        prompt = f"Question: {message} Answer:"
+        inputs = tokenizer(prompt, return_tensors="pt")
+        
+        with torch.no_grad():
+            outputs = model.generate(
+                **inputs,
+                max_length=inputs.input_ids.shape[1] + 50,
+                temperature=0.7,
+                do_sample=True,
+                pad_token_id=tokenizer.eos_token_id,
+                eos_token_id=tokenizer.eos_token_id
+            )
+        
+        full_response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+        
+        # Extract only the answer part
+        if "Answer:" in full_response:
+            answer = full_response.split("Answer:")[-1].strip()
+            return answer
+        else:
+            return full_response
+            
+    except Exception as e:
+        logger.error(f"AI model error: {e}")
+        return get_mock_response(message)
+
+async def generate_hf_response(message: str) -> str:
+    """Generate response using HuggingFace Inference API"""
+    try:
+        from huggingface_hub import InferenceClient
+        
+        # Get API key from environment
+        api_key = os.getenv("HUGGINGFACE_API_KEY")
+        if not api_key:
+            logger.warning("HUGGINGFACE_API_KEY not found, using mock response")
+            return get_mock_response(message)
+        
+        # Initialize client
+        client = InferenceClient(token=api_key)
+        
+        # Load system prompt from file or use default
+        system_prompt = load_system_prompt()
+        
+        # Create full prompt
+        full_prompt = f"<|system|>\n{system_prompt}\n<|user|>\n{message}\n<|assistant|>\n"
+        
+        # Generate response
+        response = client.text_generation(
+            full_prompt,
+            max_new_tokens=512,
+            temperature=0.7,
+            top_p=0.9,
+            top_k=40,
+            repetition_penalty=1.1,
+            stop_sequences=["<|end|>", "<|user|>"]
         )
+        
+        # Extract only the assistant's response
+        if "<|assistant|>" in response:
+            assistant_response = response.split("<|assistant|>")[-1].strip()
+            assistant_response = assistant_response.replace("<|end|>", "").strip()
+            return assistant_response
+        else:
+            return response
+            
+    except Exception as e:
+        logger.error(f"HuggingFace API error: {e}")
+        return get_mock_response(message)
+
+def get_mock_response(message: str) -> str:
+    """Fallback mock responses"""
+    mock_responses = {
+        "dimana lokasi textilindo": "Textilindo berkantor pusat di Jl. Raya Prancis No.39, Kosambi Tim., Kec. Kosambi, Kabupaten Tangerang, Banten 15213",
+        "jam berapa textilindo beroperasional": "Jam operasional Senin-Jumat 08:00-17:00, Sabtu 08:00-12:00.",
+        "jam berapa textilindo buka": "Jam operasional Senin-Jumat 08:00-17:00, Sabtu 08:00-12:00.",
+        "berapa ketentuan pembelian": "Minimal order 1 roll per jenis kain",
+        "apa ada gratis ongkir": "Gratis ongkir untuk order minimal 5 roll.",
+        "apa bisa dikirimkan sample": "Hallo kak untuk sampel kita bisa kirimkan gratis ya kak 😊"
+    }
+    
+    # Simple keyword matching
+    user_lower = message.lower()
+    for key, mock_response in mock_responses.items():
+        if any(word in user_lower for word in key.split()):
+            return mock_response
+    
+    return "Halo! Saya adalah asisten AI Textilindo. Bagaimana saya bisa membantu Anda hari ini? 😊"
+
+def load_system_prompt() -> str:
+    """Load system prompt from file or return default"""
+    try:
+        system_prompt_path = "configs/system_prompt.md"
+        if Path(system_prompt_path).exists():
+            with open(system_prompt_path, 'r', encoding='utf-8') as f:
+                content = f.read()
+            
+            # Extract SYSTEM_PROMPT from markdown if it exists
+            if 'SYSTEM_PROMPT = """' in content:
+                start = content.find('SYSTEM_PROMPT = """') + len('SYSTEM_PROMPT = """')
+                end = content.find('"""', start)
+                system_prompt = content[start:end].strip()
+            else:
+                # Use entire content
+                system_prompt = content.strip()
+            
+            return system_prompt
+        else:
+            # Default system prompt
+            return """You are a friendly and helpful AI assistant for Textilindo, a textile company.
+
+Always respond in Indonesian (Bahasa Indonesia).
+Keep responses short and direct.
+Be friendly and helpful.
+Use exact information from the knowledge base.
+The company uses yards for sales.
+Minimum purchase is 1 roll (67-70 yards)."""
+    except Exception as e:
+        logger.error(f"Error loading system prompt: {e}")
+        return """You are a friendly and helpful AI assistant for Textilindo, a textile company.
 
-# Training function
+Always respond in Indonesian (Bahasa Indonesia).
+Keep responses short and direct.
+Be friendly and helpful."""
 async def train_model_async(
     model_name: str,
     dataset_path: str,