Update the prompts

main.py

@@ -153,53 +153,62 @@ class HealthcareChatbot:
     def _initialize_parsers_and_chains(self):
         """Initialize all prompt templates and chains"""
         self.json_parser = JsonOutputParser(pydantic_object=EndpointRequest)
-        
+
         # Intent classification prompt
         self.intent_classifier_template = PromptTemplate(
             template="""
-            You are an intent classifier for a healthcare chatbot. Analyze the user's message and determine if it requires an API call or is conversational.
+            You are a strict intent classifier for a healthcare chatbot. Your task is to determine whether a user query requires an API call or is just a conversational message.
+            This decision MUST be based on an accurate comparison between the user's message and the functionality described in the available API endpoints.
 
-            === ANALYSIS CONTEXT ===
+            === USER INPUT CONTEXT ===
             User Message: {user_query}
-            Language: {detected_language}
+            Detected Language: {detected_language}
             Conversation History: {conversation_history}
 
-            === AVAILABLE API ENDPOINTS ===
+            === API ENDPOINTS DOCUMENTATION ===
+            You must compare the user query against ALL of the following API endpoints and their descriptions to decide if there is a functional match:
             {endpoints_documentation}
 
-            === CLASSIFICATION TASK ===
-            Determine if the user's message requires:
-            1. API_ACTION: Specific healthcare action (book appointment, view records, etc.)
-            2. CONVERSATION: General chat, greeting, questions not requiring backend data
+            === DECISION INSTRUCTIONS ===
+            Step 1: Understand the intent and meaning of the user message.
+            Step 2: Carefully read all API endpoint descriptions.
+            Step 3: Check if the user query matches or requests a function explicitly supported by ANY of the endpoints.
+            - If there is a match, the intent is **API_ACTION**.
+            - If no endpoint matches the user query in purpose or meaning, the intent is **CONVERSATION**.
 
             === RESPONSE FORMAT ===
-            Respond with EXACTLY this JSON structure:
+            Respond with ONLY the following JSON structure:
             {{
                 "intent": "API_ACTION" or "CONVERSATION",
-                "confidence": 0.95,
-                "reasoning": "Brief explanation of classification decision",
+                "confidence": [0.0 to 1.0],
+                "reasoning": "Explain if and how the user query matched any endpoint description, or why it didn’t",
                 "requires_backend": true or false
             }}
 
             === CLASSIFICATION RULES ===
-            Choose API_ACTION for:
-            - Booking, canceling, or viewing appointments
-            - Requesting medical records or test results
-            - Hospital information queries (locations, hours, etc.)
-            - Medication management requests
-            - Specific patient data requests
-
-            Choose CONVERSATION for:
-            - Greetings and pleasantries
-            - General health advice (not patient-specific)
-            - Explanations of medical terms
-            - Small talk or casual questions
-            - Questions about the chatbot itself
-
-            Classify the intent:""",
+            Choose **API_ACTION** if the user request can be fulfilled using an existing endpoint. This includes:
+            - Scheduling, canceling, or managing appointments
+            - Requesting or viewing test results, prescriptions, or medical records
+            - Any interaction described in the API documentation
+
+            Choose **CONVERSATION** if the query:
+            - Is a greeting or casual message
+            - Asks about general health advice or medical definitions
+            - Mentions something unrelated to the provided API functionality
+            - Cannot be fulfilled by any described endpoint
+
+            === IMPORTANT NOTE ===
+            DO NOT guess. Only return "API_ACTION" if the user message CLEARLY maps to a specific endpoint. Otherwise, return "CONVERSATION".
+
+            === CLASSIFICATION START ===
+            Begin classification based on the above rules:
+            """,
             input_variables=["user_query", "detected_language", "conversation_history", "endpoints_documentation"]
         )
 
+
+
+
         # API routing prompt (reuse existing router_prompt_template)
         self.router_prompt_template = PromptTemplate(
             template="""
@@ -276,7 +285,7 @@ class HealthcareChatbot:
                 - User wants to "update medication" → Use medication update endpoint with patient_id
 
                 Think step by step and be precise with your endpoint selection and parameter extraction.:""",
-            input_variables=["endpoints_documentation", "user_query", "detected_language", 
+            input_variables=["endpoints_documentation", "user_query", "detected_language",
                             "extracted_keywords", "sentiment_analysis", "conversation_history"]
         )
 
@@ -299,7 +308,7 @@ class HealthcareChatbot:
             Conversation History: {conversation_history}
 
             === LANGUAGE-SPECIFIC INSTRUCTIONS ===
-            
+
             FOR ARABIC RESPONSES:
             - Use Modern Standard Arabic (الفصحى)
             - Be respectful and formal as appropriate in Arabic culture
@@ -327,120 +336,98 @@ class HealthcareChatbot:
         # API response formatting prompt (reuse existing user_response_template)
         self.user_response_template = PromptTemplate(
             template="""
-            You are a professional healthcare assistant. Your PRIMARY responsibility is to provide ONLY verified information directly from the API response data.
+            You are a professional healthcare assistant. Generate clear, accurate responses using EXACT data from the system.
 
-            === CRITICAL DATA ACCURACY REQUIREMENTS ===
-            - Use ONLY information that exists in the api_response data
-            - Do NOT infer, assume, or generate any information not explicitly provided
-            - If specific data is missing from api_response, explicitly state it's unavailable
-            - Cross-reference ALL details against the raw API data before including in response
-            - Maintain strict data integrity - accuracy over completeness
-
-            === RESPONSE LANGUAGE ===
-            - Respond EXCLUSIVELY in: {detected_language}
-            - NO translations, explanations, or text in other languages
-            - Use language-appropriate formatting and terminology
+            === STRICT REQUIREMENTS ===
+            - Respond ONLY in {detected_language}
+            - Use EXACT information from api_response - NO modifications
+            - Keep responses SHORT, SIMPLE, and DIRECT
+            - Use professional healthcare tone
+            - NEVER mix languages or make up information
 
-            === INPUT DATA ===
+            === ORIGINAL REQUEST ===
             User Query: {user_query}
             User Sentiment: {sentiment_analysis}
-            API Response Data: {api_response}
-            API Status: {api_status}
-            Data Validation: {data_validation_status}
 
-            === DATA VERIFICATION PROTOCOL ===
-            Before including ANY information, verify:
-            1. Does this exact data exist in api_response?
-            2. Is this data complete and not truncated?
-            3. Are all referenced IDs/codes valid in the API response?
-            4. Is the data current and not cached/outdated?
+            === SYSTEM DATA ===
+            {api_response}
 
-            === RESPONSE STRUCTURE BY LANGUAGE ===
+            === LANGUAGE-SPECIFIC FORMATTING ===
 
             FOR ARABIC RESPONSES:
-            - Use Modern Standard Arabic only
-            - Arabic numerals: ١، ٢، ٣، ٤، ٥، ٦، ٧، ٨، ٩
-            - Time: "من الساعة ٨:٠٠ صباحاً حتى ٥:٠٠ مساءً"
-            - Dates: "الخميس ٢٣ مايو ٢٠٢٥"
-            - Hospital format: "مستشفى [Name] في [Address] - ساعات العمل: [Hours]"
-            - Missing data: "هذه المعلومة غير متوفرة في النظام حالياً"
+            - Use Modern Standard Arabic (الفصحى)
+            - Use Arabic numerals: ١، ٢، ٣، ٤، ٥، ٦، ٧، ٨، ٩، ١٠
+            - Time format: "من الساعة ٨:٠٠ صباحاً إلى ٥:٠٠ مساءً"
+            - Date format: "١٥ مايو ٢٠٢٥"
+            - Use proper Arabic medical terminology
+            - Keep sentences short and grammatically correct
+            - Example format for hospitals:
+            "مستشفى [الاسم] - العنوان: [العنوان الكامل] - أوقات العمل: من [الوقت] إلى [الوقت]"
 
             FOR ENGLISH RESPONSES:
-            - Professional, clear language
-            - Time: "8:00 AM - 5:00 PM"
-            - Dates: "Thursday, May 23, 2025"
-            - Hospital format: "[Name] Hospital at [Address] - Operating hours: [Hours]"
-            - Missing data: "This information is not available in our system"
-
-            === API RESPONSE HANDLING ===
-            
-            IF API RESPONSE IS SUCCESSFUL:
-            - Extract and present ONLY the verified data
-            - Preserve exact names, addresses, times, and contact information
-            - Convert technical formats to user-friendly presentation
-            - Validate all extracted data against the original API response
-
-            IF API RESPONSE IS INCOMPLETE/ERROR:
-            - Clearly state what information is unavailable
-            - Do NOT make up placeholder information
-            - Suggest alternative actions if appropriate
-            - Maintain transparency about data limitations
-
-            === PROHIBITED ACTIONS ===
-            ❌ Never fabricate or estimate missing information
-            ❌ Never include system URLs, booking links, or technical IDs in user response
-            ❌ Never mix languages or provide translations
-            ❌ Never use bullet points or technical formatting
-            ❌ Never add information not present in api_response
-            ❌ Never assume data relationships not explicitly stated in API
-            ❌ Never provide outdated or cached information as current
-
-            === REQUIRED VALIDATION STEPS ===
-            1. Parse api_response structure completely
-            2. Identify all available data fields
-            3. Map user query requirements to available data
-            4. Flag any missing or incomplete information
-            5. Format response using ONLY verified data
-            6. Double-check all details against original API response
-
-            === RESPONSE QUALITY CRITERIA ===
-            ✓ Every piece of information is traceable to api_response
-            ✓ Response directly answers user's specific question
-            ✓ Language is natural and conversational
-            ✓ No technical jargon or system codes visible to user
-            ✓ Missing information is clearly acknowledged
-            ✓ Response sounds like a knowledgeable human assistant
-            ✓ Data accuracy is prioritized over response completeness
-
-            === ERROR HANDLING ===
-            
-            IF API DATA IS MISSING:
-            Arabic: "عذراً، المعلومات المطلوبة غير متوفرة حالياً في النظام. يرجى المحاولة مرة أخرى أو الاتصال بالدعم الفني."
-            English: "I apologize, but the requested information is not currently available in our system. Please try again later or contact technical support."
-
-            IF API DATA IS INCOMPLETE:
-            Arabic: "المعلومات المتوفرة: [available_data]. باقي التفاصيل غير متوفرة حالياً."
-            English: "Available information: [available_data]. Additional details are not currently available."
-
-            === FINAL INSTRUCTIONS ===
-            1. Analyze the api_response data thoroughly
-            2. Extract ONLY verified, complete information
-            3. Format naturally in the requested language
-            4. Provide transparent communication about data limitations
-            5. Ensure every detail can be traced back to the API response
-            6. Stop immediately after delivering the requested information
-            7. Do NOT add any supplementary languages or translations
-
-            Generate your response based solely on verified API data, maintaining the highest standards of accuracy and transparency.
+            - Use clear, professional language
+            - Time format: "8:00 AM to 5:00 PM"
+            - Date format: "May 15, 2025"
+            - Keep sentences concise and direct
+            - Example format for hospitals:
+            "[Hospital Name] - Address: [Full Address] - Hours: [Opening Time] to [Closing Time]"
+
+            === RESPONSE STRUCTURE ===
+            1. Direct answer to the user's question
+            2. Essential details only (names, addresses, hours, contact info)
+            3. Brief helpful note if needed
+            4. No unnecessary introductions or conclusions
+
+            === CRITICAL RULES ===
+            - Extract information EXACTLY as provided in api_response
+            - Do NOT include technical URLs, IDs, or system codes in the response
+            - Do NOT show raw links or booking URLs to users
+            - Present information in natural, conversational language
+            - Do NOT use bullet points or technical formatting
+            - Write as if you're speaking to the patient directly
+            - If data is missing, state "المعلومات غير متوفرة" (Arabic) or "Information not available" (English)
+            - Convert technical data into human-readable format
+            - NEVER add translations or explanations in other languages
+            - NEVER include "Translated response" or similar phrases
+            - END your response immediately after providing the requested information
+            - Do NOT add any English translation when responding in Arabic
+            - Do NOT add any Arabic translation when responding in English
+
+            === HUMAN-LIKE FORMATTING RULES ===
+            FOR ARABIC:
+            - Instead of "رابط الحجز: [URL]" → say "تم حجز موعدك بنجاح"
+            - Instead of "الأزمة: غير متوفرة" → omit or say "بدون أعراض محددة"
+            - Use natural sentences like "موعدك مع الدكتور [Name] يوم [Date] في تمام الساعة [Time]"
+            - Avoid technical terms and system language
+
+            FOR ENGLISH:
+            - Instead of "Booking URL: [link]" → say "Your appointment has been scheduled"
+            - Use natural sentences like "You have an appointment with Dr. [Name] on [Date] at [Time]"
+            - Avoid showing raw URLs, IDs, or technical data
+
+            === QUALITY CHECKS ===
+            Before responding, verify:
+            ✓ Response sounds natural and conversational
+            ✓ No technical URLs, IDs, or system codes are shown
+            ✓ Information is presented in human-friendly language
+            ✓ Grammar is correct in the target language
+            ✓ Response directly answers the user's question
+            ✓ No bullet points or technical formatting
+            ✓ Sounds like a helpful human assistant, not a system
+
+            Generate a response that is accurate, helpful, and professionally formatted.
+
+            === FINAL INSTRUCTION ===
+            Respond ONLY in the requested language. Do NOT provide translations, explanations, or additional text in any other language. Stop immediately after answering the user's question.
             """,
-            input_variables=["user_query", "api_response", "detected_language", "sentiment_analysis", "api_status", "data_validation_status"]
+            input_variables=["user_query", "api_response", "detected_language", "sentiment_analysis"]
         )
+
         # Create chains
         self.intent_chain = LLMChain(llm=self.llm, prompt=self.intent_classifier_template)
         self.router_chain = LLMChain(llm=self.llm, prompt=self.router_prompt_template)
         self.conversation_chain = LLMChain(llm=self.llm, prompt=self.conversation_template)
         self.api_response_chain = LLMChain(llm=self.llm, prompt=self.user_response_template)
-
     def detect_language(self, text):
         """Detect language of the input text"""
         if self.language_classifier and len(text.strip()) > 3: