app.py

from huggingface_hub import InferenceClient
import os
import json
import requests
import boto3
import streamlit as st
from dotenv import load_dotenv
 
load_dotenv()
 
# Environment config
OPENWEATHERMAP_API_KEY = os.getenv("OPENWEATHERMAP_API_KEY")
AWS_REGION = os.getenv("AWS_REGION", "us-east-1")
 
# Bedrock Claude client
bedrock = boto3.client("bedrock-runtime", region_name=AWS_REGION)
 
# App UI
st.title("Weather Assistant - Umbrella Advisor")
st.markdown("Ask me if you should carry an umbrella tomorrow!")
 
# Session state for chat
if "messages" not in st.session_state:
    st.session_state.messages = []
 
# Chat history display
for msg in st.session_state.messages:
    with st.chat_message(msg["role"]):
        st.markdown(msg["content"])
 
 
def get_weather(location):
    """Get weather forecast for a specific location"""
    print(f"Getting weather for: {location}")
   
    # Validate location input
    if not location or location.strip() == "":
        return {"error": "Please specify a valid location/city name."}
   
    location = location.strip()
    geo_url = f"http://api.openweathermap.org/geo/1.0/direct?q={location}&limit=1&appid={OPENWEATHERMAP_API_KEY}"
   
    try:
        geo_resp = requests.get(geo_url).json()
       
        if not geo_resp or len(geo_resp) == 0:
            return {"error": f"Location '{location}' not found. Please check the spelling and try again."}
       
        lat, lon = geo_resp[0]['lat'], geo_resp[0]['lon']
    except (KeyError, IndexError, requests.RequestException) as e:
        return {"error": f"Error getting location data for '{location}': {str(e)}"}
    try:
        weather_url = f"http://api.openweathermap.org/data/2.5/forecast?lat={lat}&lon={lon}&appid={OPENWEATHERMAP_API_KEY}&units=metric"
        weather_data = requests.get(weather_url).json()
       
        if 'list' not in weather_data:
            return {"error": f"Unable to get weather forecast for '{location}'."}
       
        forecast = []
        for f in weather_data['list'][:8]:  # Next 24 hours
            forecast.append({
                "time": f["dt_txt"],
                "description": f["weather"][0]["description"],
                "rain_probability": f.get("pop", 0) * 100,
                "temp": f["main"]["temp"],
                "humidity": f["main"]["humidity"]
            })
       
        return {
            "location": location,
            "forecast": forecast
        }
    except (KeyError, requests.RequestException) as e:
        return {"error": f"Error getting weather forecast for '{location}': {str(e)}"}
 
 
def generate_react_response(user_input, conversation_history=""):
    """Generate response using ReAct (Reasoning + Acting) approach"""
   
    system_prompt = """You are a helpful weather assistant that uses ReAct (Reasoning + Acting) methodology to help users decide about carrying umbrellas.
 
Follow this process:
1. **Think**: Analyze what the user is asking
2. **Act**: Use available tools if needed
3. **Observe**: Process the results
4. **Reason**: Draw conclusions and provide advice
 
Available tools:
- get_weather(location): Gets weather forecast for tomorrow
 
When you need to get weather data, respond with this JSON format:
{
  "thought": "I need to get weather data for [location] to advise about umbrella",
  "action": "get_weather",
  "action_input": {"location": "city_name"}
}
 
When you have all needed information, provide a conversational response that includes:
- The location
- Your reasoning based on weather conditions
- Clear umbrella advice
 
Example: "You do not need to carry an umbrella tomorrow as the weather in New York will be sunny with no chance of rain."
 
If the user doesn't specify a location, ask them to specify it conversationally."""
 
    # Build conversation context
    messages = [
        {"role": "user", "content": f"{system_prompt}\n\nConversation history: {conversation_history}\n\nUser: {user_input}"}
    ]
   
    claude_body = {
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 1000,
        "temperature": 0.7,
        "top_p": 0.9,
        "messages": messages
    }
 
    response = bedrock.invoke_model(
        modelId="anthropic.claude-3-sonnet-20240229-v1:0",
        contentType="application/json",
        accept="application/json",
        body=json.dumps(claude_body),
    )
 
    content = json.loads(response["body"].read())["content"][0]["text"].strip()
 
    # Try to parse as ReAct JSON
    try:
        react_response = json.loads(content)
       
        if react_response.get("action") == "get_weather":
            location = react_response.get("action_input", {}).get("location", "").strip()
            thought = react_response.get("thought", "")
           
            # Validate location before calling weather function
            if not location:
                return "I need to know which city or location you're asking about. Could you please specify the location?"
           
            # Get weather data
            weather_data = get_weather(location)
           
            if "error" in weather_data:
                return weather_data["error"]
           
            # Process weather data and generate final reasoning
            try:
                reasoning_prompt = f"""Based on this weather data for {location}, provide your final umbrella recommendation:
 
Weather forecast: {json.dumps(weather_data, indent=2)}
 
Your previous thought: {thought}
 
Provide a conversational response that includes:
1. The location
2. Your reasoning based on the weather conditions
3. Clear umbrella advice
 
Format like: "You [do/do not] need to carry an umbrella tomorrow as the weather in [location] will be [conditions and reasoning]."
"""
               
                final_messages = [{"role": "user", "content": reasoning_prompt}]
               
                final_body = {
                    "anthropic_version": "bedrock-2023-05-31",
                    "max_tokens": 500,
                    "temperature": 0.7,
                    "messages": final_messages
                }
               
                final_response = bedrock.invoke_model(
                    modelId="anthropic.claude-3-sonnet-20240229-v1:0",
                    contentType="application/json",
                    accept="application/json",
                    body=json.dumps(final_body),
                )
               
                final_content = json.loads(final_response["body"].read())["content"][0]["text"].strip()
                return final_content
            except Exception as e:
                return f"Error processing weather data: {str(e)}"
           
    except json.JSONDecodeError:
        # If not JSON, return the content as is (probably asking for location)
        pass
   
    return content
 
 
def build_conversation_history():
    """Build conversation history for context"""
    history = []
    for msg in st.session_state.messages[-4:]:  # Last 4 messages for context
        history.append(f"{msg['role'].capitalize()}: {msg['content']}")
    return "\n".join(history)
 
 
if prompt := st.chat_input("Type your question here..."):
    st.session_state.messages.append({"role": "user", "content": prompt})
    with st.chat_message("user"):
        st.markdown(prompt)
 
    with st.chat_message("assistant"):
        with st.spinner("Thinking..."):
            conversation_history = build_conversation_history()
            reply = generate_react_response(prompt, conversation_history)
            st.markdown(reply)
    st.session_state.messages.append({"role": "assistant", "content": reply})
 
 
with st.sidebar:
    st.header("About")
    st.markdown("""
    - Powered by **AWS Bedrock (Claude Sonnet)**  
    - Uses **ReAct (Reasoning + Acting)** methodology
    - Retrieves real-time data from **OpenWeatherMap**  
    - Provides step-by-step reasoning for umbrella advice
    """)
   
    st.subheader("Sample Prompts")
    st.markdown("""
    - Should I bring an umbrella tomorrow?  
    - Will it rain in Delhi tomorrow?  
    - Do I need an umbrella in Tokyo?
    - Should I carry an umbrella tomorrow in London?
    """)
   
    st.subheader("ReAct Process")
    st.markdown("""
    1. **Think**: Analyze your question
    2. **Act**: Get weather data if needed
    3. **Observe**: Process weather information  
    4. **Reason**: Provide umbrella advice with explanation
    """)