Example of the chatbot's response in the Streamlit web UI.
Graph structure of the conversation pipeline implemented with LangGraph.
This project is an intelligent chatbot that helps users discover the best outdoor trails based on their location, preferred activities, and desired trail features, while also considering real-time weather conditions. By integrating trail data, Google Maps, and weather APIs, the chatbot provides personalized recommendations for outdoor adventures, ensuring both suitability and safety.
- Install dependencies:
pip install -r requirements.txt
- Set up your
.envfile with your API keys:OPENWEATHER_API_KEY=your_openweather_key GOOGLE_API_KEY=your_google_maps_key
- Run the Streamlit app:
streamlit run st_app.py
User: I'm in Boston, MA and want to go hiking and see a waterfall.
Assistant: [Recommends best trail, provides weather, and suggests alternatives.]
The main goal is to assist users in finding the most suitable trails for their outdoor activities, factoring in:
- Proximity to the user's location
- Supported activities (e.g., hiking, biking, birding)
- Desired features (e.g., waterfalls, lakes, dog-friendly)
- Current weather conditions at the trailhead
The chatbot follows a multi-step pipeline:
- Parse User Input: Extracts the user's location, activities, and desired features from natural language input.
- Find Trails: Retrieves and filters nearby trails using a trail database and Google Maps for distance calculations.
- Rank & Select Trail: Ranks trails based on best match for user preferences and proximity.
- Check Weather: Fetches real-time weather data for the selected trail.
- Evaluate Weather: Determines if the weather is suitable for the intended activity.
- Compose Response: Crafts a friendly, informative reply with the best match and alternative suggestions.
The core logic is implemented as a directed graph using LangGraph, where each node represents a step in the conversation flow:
- Classifier: Determines if the message should be parsed for trail search or handled as a general query.
- Router: Directs the flow based on message type.
- ParseUserInput: Extracts structured data from user input.
- FindTrails: Retrieves and ranks trails.
- SelectTrail: Picks the best candidate.
- CheckWeather: Gets weather for the trail.
- EvaluateWeather: Decides if the weather is suitable.
- IncrementIndex: Moves to the next trail if needed.
- ComposeResponse: Generates the final reply.
- Natural Language Understanding: Uses LLMs (Anthropic Claude) to parse user intent and classify messages.
- Data Integration: Combines trail data, Google Maps API (for geocoding and distance), and OpenWeather API.
- LLM-based DataFrame Agent: Uses LangChain's
create_pandas_dataframe_agentto rank and select trails based on user preferences and proximity. - Conversational Graph: Implements a stateful, multi-step conversation using LangGraph.
- Pandas DataFrame Manipulation: For efficient trail filtering and ranking.
- Python 3.10+
- Streamlit: For the interactive web UI
- LangGraph: For graph-based conversational flow
- LangChain: For LLM orchestration and DataFrame agent
- Anthropic Claude: For language understanding and response generation
- Google Maps API: For geocoding and distance calculations
- OpenWeather API: For real-time weather data
- Pydantic: For data validation
- Outdoor Enthusiasts: Find the best trails for hiking, biking, birding, and more, tailored to current weather.
- Trip Planning: Plan safe and enjoyable outdoor activities based on real-time conditions.
- Accessibility: Discover trails with specific features (e.g., ADA accessible, dog-friendly, kid-friendly).
- Conversational Assistant: Integrate as a chatbot in web or mobile apps for personalized outdoor recommendations.
For more details, see the code in trail_weather_graph.py and the Streamlit UI in st_app.py.