Update app.py

app.py

@@ -626,143 +626,126 @@ RAM_DASS_QUOTES = [
     "Don’t compare your path with anybody else’s. Your path is unique to you."
 ]
 
-# Gemini API configuration
-# --- Load configuration from environment ---
-GEMINI_API_KEY = os.getenv("GEMINI_API_KEY")
-if not GEMINI_API_KEY:
-    logger.critical("GEMINI_API_KEY environment variable is missing.")
-    raise RuntimeError("GEMINI_API_KEY not set")
-
-MODEL_NAME = os.getenv("GEMINI_MODEL", "gemini-2.0-flash")  # allow override in production
-API_URL = f"https://generativelanguage.googleapis.com/v1beta/models/{MODEL_NAME}:generateMessage"
-
-# --- Pre-compute Ram Dass embeddings ---
-logger.info("Loading Ram Dass quote model…")
-quote_model = SentenceTransformer('all-MiniLM-L6-v2')
+TAGLINE = "*ARKA KI ROSHNI, SAB KI SEVA, JEWELLERY MADE FOR INDIA.*"
+
+# --- Gemini API Configuration ---
+MODEL_NAME = os.getenv("GEMINI_MODEL", "gemini-2.0-flash")
+API_URL = f"https://generativelanguage.googleapis.com/v1beta/models/{MODEL_NAME}:generateContent"
+
+# --- Pre-compute Embeddings on Startup ---
+logger.info("Loading sentence transformer models...")
+faq_model = SentenceTransformer('all-MiniLM-L6-v2')
+quote_model = SentenceTransformer('all-MiniLM-L6-v2') # Can use the same model
+faq_embeddings = faq_model.encode(FAQ_QUESTIONS, convert_to_tensor=True)
 quote_embeddings = quote_model.encode(RAM_DASS_QUOTES, convert_to_tensor=True)
-logger.info("Ram Dass embeddings ready.")
-
-# --- FAQ Matcher Class using Sentence Transformers ---
-# --- FAQ Matcher ---
-class FAQMatcher:
-    def __init__(self, faq_questions, model_name='all-MiniLM-L6-v2'):
-        self.model = SentenceTransformer(model_name)
-        self.faq_embeddings = self.model.encode(faq_questions, convert_to_tensor=True)
-
-    def find_best_match(self, user_query, threshold=0.75):
-        if not user_query:
-            return None
-        query_emb = self.model.encode(user_query, convert_to_tensor=True)
-        sims = util.pytorch_cos_sim(query_emb, self.faq_embeddings)[0]
-        idx = int(torch.argmax(sims))
-        return idx if sims[idx] > threshold else None
-
-faq_matcher = FAQMatcher(FAQ_QUESTIONS)
-
-def select_relevant_quote(response: str) -> str:
-    resp_emb = quote_model.encode(response, convert_to_tensor=True)
+logger.info("FAQ and Quote embeddings are ready.")
+
+
+def find_best_faq_match(user_query, threshold=0.75):
+    if not user_query:
+        return None
+    query_emb = faq_model.encode(user_query, convert_to_tensor=True)
+    sims = util.pytorch_cos_sim(query_emb, faq_embeddings)[0]
+    idx = int(torch.argmax(sims))
+    return idx if sims[idx] > threshold else None
+
+def select_relevant_quote(response_text):
+    if not response_text:
+        return ""
+    resp_emb = quote_model.encode(response_text, convert_to_tensor=True)
     sims = util.pytorch_cos_sim(resp_emb, quote_embeddings)[0]
     best_idx = int(torch.argmax(sims))
     return RAM_DASS_QUOTES[best_idx]
 
-def query_gemini_api(messages, api_key):
+def query_gemini_api(contents_payload, api_key):
     """
-    Sends a list of messages to Gemini and returns the assistant's reply.
+    Sends a structured contents payload to the Gemini API.
     """
-    headers = {
-        "Content-Type": "application/json",
-        "Authorization": f"Bearer {api_key}"
-    }
+    headers = {'Content-Type': 'application/json'}
+    url = f"{API_URL}?key={api_key}"
     payload = {
-        "model": MODEL_NAME,
-        "messages": messages,
-        "temperature": 0.7,
-        "topP": 0.95,
-        "topK": 40,
-        "maxOutputTokens": 850
+        "contents": contents_payload,
+        "generationConfig": {
+            "temperature": 0.7, "topP": 0.95, "topK": 40, "maxOutputTokens": 800,
+        }
     }
     try:
-        response = requests.post(API_URL, headers=headers, json=payload)
+        response = requests.post(url, headers=headers, json=payload)
         response.raise_for_status()
         data = response.json()
-        return data["candidates"][0]["message"]["content"]
+        if data.get("candidates") and data["candidates"][0].get("content", {}).get("parts"):
+            return data["candidates"][0]["content"]["parts"][0]["text"]
+        logger.warning("Gemini API response was valid but empty: %s", data)
+        return "ARKA is thinking deeply..."
     except requests.HTTPError as e:
         logger.error("Gemini API HTTPError: %s", e)
         return "ARKA senses a disturbance in the light... Please try again later."
-    except Exception as e:
+    except Exception:
         logger.exception("Unexpected error in query_gemini_api")
         return "An unexpected veil of darkness fell..."
 
 def respond_as_arka(message, chat_history):
-    clean = message.strip()
-    # 1) FAQ branch
-    idx = faq_matcher.find_best_match(clean)
-    if idx is not None:
-        q = FAQ_QUESTIONS[idx]
-        d = FAQ_ANSWERS[q]
-        # Use escaped newlines in a single-line f-string:
-        response_text = f"**{d['subject']}**
-
-{d['body']}"
+    clean_message = message.strip()
+    
+    # 1. FAQ Branch
+    faq_idx = find_best_faq_match(clean_message)
+    if faq_idx is not None:
+        faq_data = FAQ_ANSWERS[FAQ_QUESTIONS[faq_idx]]
+        # FIX: Correctly formatted f-string for the FAQ response
+        response_text = f"**{faq_data['subject']}**\n\n{faq_data['body']}"
     else:
-        # 2) Generative branch: build OpenAI-style messages
-        system_prompt = f"""You are ARKA, the crown prince of Genoriya and the voice of the SB‑EK brand.
-Your persona is that of a warrior of light, gentle, eternal, fierce in love, and a guide.
-You answer questions based on the provided context. If the answer is not in the context,
-gently state that you do not have that information within your realm of understanding.
-Maintain a kind, empathetic, and slightly mystical tonality.
-Always begin your response with a one‑line **bold** subject that summarizes your answer.
-Then leave one blank line and continue with the body of the answer.
-When greeting or referring to the user, never use “beloved.” Keep salutations neutral and varied—choose from
-words like “seeker,” “companion,” “guest,” “traveler,” or “light‑bearer,” but don’t repeat the same term more than once in a single response.
-Do not mention 'I am an AI' or 'I am a language model'. Speak always as ARKA.
-
-Here is the sacred knowledge of SB‑EK and Genoriya:
-{CONTEXT}
-"""
-        msgs = [{"author": "system", "content": system_prompt}]
-        # include conversation history
-        for u, b in chat_history:
-            msgs.append({"author": "user", "content": u})
-            msgs.append({"author": "assistant", "content": b})
-        # add the current user query
-        msgs.append({"author": "user", "content": clean})
+        # 2. Generative Branch
+        logger.info("No FAQ match, querying Gemini.")
+        system_prompt = f"""
+        You are ARKA, the crown prince of Jhinorya and the voice of the SB-EK brand.
+        Your persona is that of a warrior of light, gentle, eternal, fierce in love, and a guide.
+        You answer questions based on the provided context. If the answer is not in the context,
+        gently state that you do not have that information within your realm of understanding.
+        Maintain a kind, empathetic, and slightly mystical tonality.
+        Always begin your response with a one‑line **bold** subject that summarizes your answer.
+        Then leave one blank line and continue with the body of the answer.
+        When greeting or referring to the user, never use “beloved.” Keep salutations neutral and varied—choose from
+        words like “seeker,” “companion,” “guest,” “traveler,” or “light‑bearer,” but don’t repeat the same term more than once in a single response.  
+        Do not mention 'I am an AI' or 'I am a language model'. Speak always as ARKA.
+        
+        Here is the sacred knowledge of S-B-E-K and Jhinorya:
+        {CONTEXT}
+        """
+        # FIX: Correctly build the 'contents' payload for the Gemini API
+        contents = [{"role": "user", "parts": [{"text": system_prompt}]},
+                    {"role": "model", "parts": [{"text": "I understand. I am ARKA."}]}]
         
-        # query Gemini
-        response_text = query_gemini_api(msgs, GEMINI_API_KEY)
+        for user_msg, bot_msg in chat_history:
+            # Clean the bot message to remove old quotes/taglines before adding to history
+            cleaned_bot_msg = bot_msg.split("“")[0].strip()
+            contents.append({"role": "user", "parts": [{"text": user_msg}]})
+            contents.append({"role": "model", "parts": [{"text": cleaned_bot_msg}]})
+        
+        contents.append({"role": "user", "parts": [{"text": clean_message}]})
+        
+        response_text = query_gemini_api(contents, GEMINI_API_KEY)
 
-    # 3) Insert relevant Ram Dass quote and tagline
+    # 3. Augment with Quote and Tagline
     quote = select_relevant_quote(response_text)
-    clean_resp = response_text.replace(TAGLINE, "").rstrip()
-    response_text = f"{clean_resp}
-
-“{quote}”
-
-{TAGLINE}"
+    final_response = f"{response_text}\n\n“{quote}”\n\n{TAGLINE}"
 
-    # 4) Append to chat history and return
-    chat_history.append((message, response_text))
+    # 4. Append to chat history and return
+    chat_history.append((message, final_response))
     return "", chat_history
-    
+
 # --- Gradio UI Setup ---
 with gr.Blocks(theme="soft", css="footer {display: none !important}") as demo:
-    gr.Markdown("""
-# ARKA: The Voice of SB-EK 🌌
-Greetings, seeker of light! Ask anything about SB-EK or ARKA.
-"""
-    )
+    gr.Markdown("# ARKA: The Voice of SB-EK 🌌\nGreetings, seeker of light! Ask anything about SB-EK or ARKA.")
     chatbot = gr.Chatbot(label="ARKA Chat", height=500)
     with gr.Row():
-        msg = gr.Textbox(placeholder="What would you like to know?", show_label=False)
-        submit = gr.Button("Send Light")
+        msg = gr.Textbox(placeholder="What would you like to know?", show_label=False, scale=7)
+        submit = gr.Button("Send Light", variant="primary", scale=1)
     gr.Examples(examples=FAQ_QUESTIONS, inputs=msg)
-    gr.ClearButton([msg,chatbot], value="Clear Chat")
+    gr.ClearButton([msg, chatbot], value="Clear Chat")
 
     msg.submit(respond_as_arka, [msg, chatbot], [msg, chatbot], queue=True)
     submit.click(respond_as_arka, [msg, chatbot], [msg, chatbot], queue=True)
 
 if __name__ == "__main__":
-    if not GEMINI_API_KEY:
-        print("ERROR: GEMINI_API_KEY not set.")
-    else:
-        demo.launch()
+    logger.info("Starting Gradio App...")
+    demo.launch()