🌿 MangroveGPT: Real-Time Multi-Modal Multi-Agent Mangrove Monitoring with Agentic RAG

MangroveGPT is an AI-powered system that monitors mangrove health in real time by integrating multi-modal Earth observation data. It combines time series forecasting, satellite imagery analysis, and natural language understanding to support environmental research and decision-making.

🚀 Key Features

• Real-Time Forecasting: Predicts mangrove health using environmental station data and an XGBoost time series model trained on NDVI (Normalized Difference Vegetation Index).
• Multi-Modal Query Routing: Automatically classifies and routes textual and image-based user queries.
• Satellite Image Understanding: Uses a fine-tuned CLIP vision-language model and captioning pipeline to describe satellite images and route them appropriately.
• RAG Pipeline: Supports retrieval-augmented generation (RAG) using OpenAI + Pinecone + LangChain to answer research-based questions.
• Streamlit App: Deployed with an easy-to-use web interface for users from any background—no coding required.

---

🧠 How It Works

1. User Query → Interpreted using a structured LLM.
2. Query Type Classification:
   • Text queries → Forecast or Research pipeline
   • Image queries → CLIP-based captioning → Text pipeline
3. Forecast Pipeline:
   • Fetches real-time data (wind, water level, NDVI)
   • Builds a 7-week lagged feature vector
   • Predicts NDVI using XGBoost
   • Summarizes with LLM
4. Research Pipeline:
   • Uses LangChain and Pinecone to answer general mangrove questions
5. Interface:
   • Accessible through Streamlit for real-time usage.

---

🛠 Tech Stack

• LLMs: OpenAI (ChatGPT / Gemini), LangChain
• Data APIs: NOAA, Google Earth Engine, Google Maps
• ML: XGBoost (time series forecasting)
• Vision: CLIP (fine-tuned on mangrove imagery)
• Search: Pinecone
• Frontend: Streamlit
• Framework: LangGraph (multi-agent pipeline)

---

📊 Model Performance

• MAPE: 2% (Mean Absolute Percentage Error)
• Roughly equivalent to 98% accuracy for NDVI forecasting — extremely reliable for environmental insights.

---

💻 Getting Started

# Clone the repo
git clone https://github.com/GilbertHarijanto/MangroveAgent.git
cd mangrovegpt

# Install dependencies
pip install -r requirements.txt

# Launch FastAPI backend
uvicorn backend.main:app --reload

# In a separate terminal, launch React frontend
cd frontend
npm install
npm run dev

# Optional: legacy Streamlit app
streamlit run app.py