README.md

metadata

title: Odisha Disaster RAG Chatbot
emoji: 🌊
colorFrom: green
colorTo: yellow
sdk: docker
app_file: app.py
pinned: false
license: mit

🌀 Odisha Disaster Management RAG Chatbot

📌 Overview

Odisha faces recurring disasters every year such as floods, cyclones, and droughts.
While the state has a strong disaster management authority (OSDMA), information is often scattered across reports, research papers, and government documents.

This project builds a Retrieval-Augmented Generation (RAG) based chatbot that provides citizens, researchers, and policymakers with clear, reliable, and contextual answers related to Odisha’s disaster management practices.

---

✨ Features

• Handles 132 PDFs and 12 text files (OSDMA, IMD, NDMA, research papers).
• Preprocessing pipeline: PDF/text extraction, cleaning, normalization, chunking.
• Embeddings with sentence-transformers/all-MiniLM-L6-v2.
• FAISS Vector Database for fast and efficient retrieval.
• RAG pipeline:

1. User query → query structuring (handles poor English, spelling issues).
2. Retrieve relevant chunks from FAISS.
3. If no relevant results → no LLM call (saves cost).
4. If relevant → LLM generates structured, contextual answers.

• Prompt engineering for better accuracy and reduced hallucinations.
• Backend: FastAPI.
• Frontend: HTML, CSS, JS chatbot interface.

---

🏗️ Architecture

User Query → Query Structuring → FAISS Retriever → Relevant Chunks → LLM → Answer

🛠️ Tech Stack

• Python (data handling & backend)
• PyPDF, TextLoader → PDF/Text extraction
• FAISS → Vector database
• HuggingFace Sentence Transformers → Embeddings
• FastAPI → Backend API
• HTML, CSS, JavaScript → Frontend chatbot UI
• LLM (OpenAI / HuggingFace) → Answer generation

---

⚙️ Installation

1. Clone the repository

git clone https://github.com/subhakanta156/odisha-disaster-knowledge-assistant.git

2. Create virtual environment & install dependencies

python -m venv venv
source venv/bin/activate   # Linux/Mac
venv\Scripts\activate      # Windows

pip install -r requirements.txt

3. Prepare the data

• Place all PDFs/text files inside the data/ folder.
• Run preprocessing & embedding script:

python scripts/build_vector_store.py

4. Run the FastAPI backend

uvicorn app.main:app --reload

5. Open the frontend

• Open frontend/index.html in your browser.

🚀 Usage

Ask questions like:

• “How does Odisha’s disaster proneness compare with other Indian states?”
• “Provide details of relief funds sanctioned for Odisha during the 1999 Super Cyclone.”
• “Which Odisha agency is primarily responsible for issuing cyclone alerts?”
• “Explain the key steps taken by the Odisha government if lives are lost in a disaster?”

The system retrieves relevant chunks from reports and generates reliable, structured answers.

---

📊 Optimizations

• Added query filtering → No LLM call if retrieval fails (reduces cost).
• Handled poor English queries via query restructuring.
• Improved prompt engineering to minimize hallucinations.

---

📌 Future Improvements

• Add multilingual support (Odia/Hindi queries).
• Deploy on cloud (AWS/GCP/Azure) with Docker.
• Use advanced embeddings (e.g., all-mpnet-base-v2) for higher accuracy.
• Add real-time updates (e.g., cyclone alerts).

---

👨‍💻 Author

Subhakanta Rath

MSc AI & ML @ IIIT Lucknow

Passionate about AI/ML, Data Engineering