Update app.py

app.py

@@ -53,7 +53,14 @@ from crewai.tools import tool
 from geopy.geocoders import Nominatim
 from timezonefinder import TimezoneFinder
 from langchain_experimental.agents import create_pandas_dataframe_agent
+import pinecone
 
+pinecone.init(
+    api_key=os.environ["PINECONE_API_KEY"],
+    environment=os.environ["PINECONE_ENVIRONMENT"]
+)
+
+index = pinecone.Index("rag-docs")
 
 session_retriever = None
 session_qa_chain = None
@@ -293,7 +300,7 @@ def search_web(query):
     api_key = os.environ.get("SERPAPI_API_KEY")
     if not api_key:
         return "SERPAPI_API_KEY not set. Please set the environment variable."
-    params = {"engine": "google", "q": query, "api_key": api_key, "num": 5}
+    params = {"engine": "google", "q": query, "api_key": api_key, "num": 10}
     search = GoogleSearch(params)
     results = search.get_dict()
     if "organic_results" in results:
@@ -303,7 +310,25 @@ def search_web(query):
             link = result.get("link", "No Link")
             snippet = result.get("snippet", "No Snippet")
             raw_output += f"Title: {title}\nLink: {link}\nSnippet: {snippet}\n\n"
-        prompt = "Summarize the following search results in a concise, human-friendly way:\n" + raw_output
+        prompt = f"""
+You are a helpful assistant. Given the following web search results and the user's question:
+
+1. First, identify **only** the most relevant entries.
+2. Then, summarise the key insights using fluent, **British English**.
+3. If the user's question asks for trends, categories, or multiple items, you may present key points in a non-markdown bullet-point style (e.g. use "•" instead of "-", avoid using "**" for bold).
+4. Otherwise, reply in a short, natural paragraph.
+5. Do **not** use any markdown formatting such as `**`, `##`, or list syntax.
+6. Keep your answer concise and professional, as if explaining to a colleague.
+7. If no relevant info is found, say: "Sorry, I couldn't find a reliable answer from the current results."
+
+--- Web Search Results ---
+{raw_output}
+
+--- User's Question ---
+"{query}"
+
+Answer:
+"""
         summarized = _general_chat(prompt)
         return summarized if summarized else raw_output.strip()
     else:
@@ -618,16 +643,41 @@ def weather_agent_tool(query: str) -> str:
         hour_str = closest_hour["time"].split(" ")[1]
 
         summary_prompt = f"""
-        Summarise this weather forecast naturally:
-        Location: {location}
-        Time: {target_dt.strftime('%Y-%m-%d')} at {hour_str}
-        Condition: {condition}
-        Temp: {temp}°C (Feels like {feels}°C)
-        Humidity: {humidity}%
-        Chance of rain: {chance_rain}%
-
-        Make it short, friendly, and human-style.
-        """
+You are a helpful weather reasoning assistant.
+
+The user wants to know about the weather conditions at a specific time: {target_dt.strftime('%Y-%m-%d %H:%M')} in {location}.  
+Use the data below to answer their question. This may refer to the past, present, or future — do not assume it is the current weather.
+
+Based on the following weather data and the user's question, think step-by-step to extract the most relevant information, and give a natural, friendly, and cautious answer in British English.
+
+Avoid being overly confident — never say "Yes, it will..." or "Definitely." Instead, use expressions like:
+- "It is very likely that..."
+- "There is a high chance of..."
+- "Based on the available data, it seems that..."
+- "There may be..."
+
+Also, after answering the question, include a short weather summary and a useful suggestion (e.g., bring an umbrella, wear sunscreen, avoid outdoor activities).
+
+**Do not use markdown formatting such as `*`, `**`, or list symbols.**
+
+--- Weather Data ---
+Location: {location}
+Time: {target_dt.strftime('%Y-%m-%d')} at {hour_str}
+Condition: {condition}
+Temperature: {temp}°C (Feels like {feels}°C)
+Humidity: {humidity}%
+Chance of rain: {chance_rain}%
+Chance of snow: {closest_hour.get("chance_of_snow", "N/A")}%
+Wind speed: {closest_hour.get("wind_kph", "N/A")} kph
+UV index: {closest_hour.get("uv", "N/A")}
+Cloud cover: {closest_hour.get("cloud", "N/A")}%
+Visibility: {closest_hour.get("vis_km", "N/A")} km
+
+--- User Question ---
+{query}
+
+--- Final Answer ---
+"""
         response = llm_gpt4.invoke(summary_prompt)
         return response.content.strip() if isinstance(response, AIMessage) else str(response)
 
@@ -715,16 +765,41 @@ def weather_tool(query: str) -> str:
         hour_str = closest_hour["time"].split(" ")[1]
 
         summary_prompt = f"""
-        Summarise this weather forecast naturally:
-        Location: {location}
-        Time: {target_dt.strftime('%Y-%m-%d')} at {hour_str}
-        Condition: {condition}
-        Temp: {temp}°C (Feels like {feels}°C)
-        Humidity: {humidity}%
-        Chance of rain: {chance_rain}%
-
-        Make it short, friendly, and human-style.
-        """
+You are a helpful weather reasoning assistant.
+
+The user wants to know about the weather conditions at a specific time: {target_dt.strftime('%Y-%m-%d %H:%M')} in {location}.  
+Use the data below to answer their question. This may refer to the past, present, or future — do not assume it is the current weather.
+
+Based on the following weather data and the user's question, think step-by-step to extract the most relevant information, and give a natural, friendly, and cautious answer in British English.
+
+Avoid being overly confident — never say "Yes, it will..." or "Definitely." Instead, use expressions like:
+- "It is very likely that..."
+- "There is a high chance of..."
+- "Based on the available data, it seems that..."
+- "There may be..."
+
+Also, after answering the question, include a short weather summary and a useful suggestion (e.g., bring an umbrella, wear sunscreen, avoid outdoor activities).
+
+**Do not use markdown formatting such as `*`, `**`, or list symbols.**
+
+--- Weather Data ---
+Location: {location}
+Time: {target_dt.strftime('%Y-%m-%d')} at {hour_str}
+Condition: {condition}
+Temperature: {temp}°C (Feels like {feels}°C)
+Humidity: {humidity}%
+Chance of rain: {chance_rain}%
+Chance of snow: {closest_hour.get("chance_of_snow", "N/A")}%
+Wind speed: {closest_hour.get("wind_kph", "N/A")} kph
+UV index: {closest_hour.get("uv", "N/A")}
+Cloud cover: {closest_hour.get("cloud", "N/A")}%
+Visibility: {closest_hour.get("vis_km", "N/A")} km
+
+--- User Question ---
+{query}
+
+--- Final Answer ---
+"""
         response = llm_gpt4.invoke(summary_prompt)
         return response.content.strip() if isinstance(response, AIMessage) else str(response)
 
@@ -927,7 +1002,7 @@ def multi_agent_chat_advanced(query: str, file=None) -> str:
     date_keywords = ["what date", "today", "what time", "what day", "current time", "date", "現在幾點", "今天幾號", "禮拜幾"]
     if any(k in lower_query for k in date_keywords):
         return get_time_tool(query)
-    weather_keywords = ["weather", "rain", "snow", "cold", "hot", "cloudy", "sunny", "temperature", "forecast", "天氣", "會不會下雨", "冷嗎", "熱嗎", "氣溫"]
+    weather_keywords = ["weather", "rain", "snow", "cold", "hot", "sunscreen", "sunglasses", "umbrella", "windy", "cloudy", "sunny", "temperature", "forecast", "天氣", "會不會下雨", "冷嗎", "熱嗎", "氣溫"]
     if any(k in lower_query for k in weather_keywords):
         return weather_agent_tool(query)
     search_keywords = ["latest", "news", "startup", "startups", "company", "companies", "top", "trending", "in 2025", "in 2024", "tell me"]
@@ -1033,6 +1108,8 @@ If no relevant information is found in the document, the system will say "No rel
 
 *Note: The LLaMA module generates responses based solely on the current query without follow-up memory or chat history management.*
 
+> This is a CPU-only demo running a **quantised 1B LLaMA model**, built to show that full RAG + multi-agent systems can run even without a GPU. In production, the model can be replaced with larger ones (3B, 7B, etc.) and served using vLLM, 4-bit quantisation, or TensorRT for better speed. The design focuses on portability, deployment, and modularity.
+
 Feel free to ask any question related to Biden’s 2023 State of the Union Address.
 """
 demo_description2 = """
@@ -1058,7 +1135,15 @@ demo_description3 = """
 Upload a PDF, TXT, or DOCX file and ask a question about its content.
 This demo uses **GPT-4o-Mini** to answer questions based on the content of your uploaded document.
 
-Feel free to ask any question related to your document.
+Note: This is a **strict RAG-based QA** system. It will only answer questions if the answer is explicitly found in the uploaded document.
+For more flexible or general-purpose responses, please try Tab 1 (Multi-Agent Assistant).
+
+Typical Use Cases:
+- Legal, technical, or academic documents where factual precision is critical
+- Internal company reports where hallucination must be avoided
+- Medical papers where only referenced content should be discussed
+
+Feel free to ask any question directly related to your document.
 """
 demo_description4 = """
 **Context**:
@@ -1083,7 +1168,7 @@ This demo presents a GPT-style Multi-Agent AI Assistant, built with **LangChain,
 2. What are the key ideas mentioned in this file? *(→ RAG QA Agent)*
 3. What is LangChain used for? | What are the latest trends in AI startups in 2025? | Tell me the most recent breakthrough in quantum computing. *(→ Online Rag Agent)*
 4. What's the current time in London? | What’s today’s date in New York? | What time is it in Taipei right now? *(→ Time Agent)*
-5. Will it rain in New York in 2 hours? | Is it going to be hot tomorrow in Nottingham? | What was the weather like in Paris two days ago? | Is it gonna rain later? *(→ Weather Agent)*
+5. Will it rain or snow in Sapporo tomorrow night? | Is it too windy for cycling in Amsterdam at 6am? | Do I need to bring an umbrella later this evening in Edinburgh? | Should I wear sunscreen in Bangkok around noon tomorrow? | Is it gonna rain later? | What was the weather like in Paris on last Sunday? | Will the weather be suitable for hiking at 3pm in Lake District? *(→ Weather Agent)*
 6. If I earn $15 per hour and work 8 hours a day for 5 days, how much will I earn? | What is 5 * 22.5 / sin(45) | 3^3 + 4^2 | Calculate 25 * log(1000) | What is the square root of 144 *(→ Math Agent)*
 7. Who are you? | What can you do? | What is the meaning of life? *(→ General Chat Agent)*