[NOTICKET] Tuning prompt generate audio text

src/agents/chatbot.py

@@ -6,7 +6,6 @@ from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
 from langchain_core.output_parsers import StrOutputParser
 from src.config.settings import settings
 from src.middlewares.logging import get_logger
-from langchain_core.messages import HumanMessage, AIMessage
 
 logger = get_logger("chatbot")
 
@@ -77,21 +76,72 @@ class ChatbotAgent:
             logger.error("Response generation failed", error=str(e))
             raise
 
+    def language_hint(self, full_response: str) -> str:
+        text = full_response.lower()
+        words = set(re.findall(r"\b[\w']+\b", text, flags=re.UNICODE))
+
+        indo_markers = {
+            "yang", "dan", "untuk", "tidak", "akan", "saya", "kamu", "kita",
+            "mereka", "adalah", "ini", "itu", "dengan", "karena", "sebagai",
+            "oleh", "pada", "dari", "ke", "di"
+        }
+
+        eng_markers = {
+            "the", "and", "for", "you", "your", "i", "is", "are", "will",
+            "not", "this", "that", "with", "because", "as", "from", "to",
+            "in", "of"
+        }
+
+        indo_count = sum(1 for w in words if w in indo_markers)
+        eng_count = sum(1 for w in words if w in eng_markers)
+
+        if indo_count > eng_count and indo_count >= 2:
+            return "Indonesian"
+        if eng_count > indo_count and eng_count >= 2:
+            return "English"
+
+        if indo_count > 0 and eng_count == 0:
+            return "Indonesian"
+        if eng_count > 0 and indo_count == 0:
+            return "English"
+
+        return "the same language as the response"
+
     async def generate_audio_text(self, full_response: str) -> str:
         """Generate a 2-3 sentence TTS-friendly summary of the assistant response."""
         try:
+            lang = self.language_hint(full_response)
             prompt = (
-                "You are a text-to-speech assistant. Given the following AI response, "
-                "write a plain-language summary in 2 to 3 sentences maximum. "
-                "Rules: no markdown, no bullet points, no headers, no code, no special characters. "
-                "Write as if speaking aloud. Be concise. "
-                "IMPORTANT: use the exact same language as the response below.\n\n"
+                "You are a text to speech assistant. Given the following AI response, "
+                "write a plain language summary in exactly 2 or 3 sentences. "
+                "Output only the summary text. Allowed characters are letters numbers spaces and periods only. "
+                "Do not output any other characters. Do not name symbols. "
+                f"The response language is {lang}. Write the summary in {lang} only. Do not translate.\n\n"
                 f"Response:\n{full_response}\n\n"
-                "Summary (2-3 sentences only):"
+                "Summary:"
             )
+
             result = await self.llm.ainvoke(prompt)
-            sentences = re.split(r'(?<=[.!?])\s+', result.content.strip())
-            return " ".join(sentences[:3])
+            logger.info(f"Generated audio text: {str(result)[:250]}...")
+
+            text = result.content if hasattr(result, "content") else str(result)
+            text = text.replace("!", ".").replace("?", ".").replace(";", ".").replace("*", "")
+
+            sentences = [s.strip() for s in text.split(".") if s.strip()][:3]
+
+            def sanitize(sentence: str) -> str:
+                sentence = re.sub(r"[^A-Za-z0-9 .]", " ", sentence)
+                sentence = re.sub(r"\s+", " ", sentence).strip()
+                return sentence
+
+            sanitized = [sanitize(s) for s in sentences if s]
+            if not sanitized:
+                return ""
+
+            output = ". ".join(sanitized).strip()
+            if output and not output.endswith("."):
+                output += "."
+            return output
         except Exception as e:
             logger.error("Audio text generation failed", error=str(e))
             return ""