J.A.R.V.I.S: GeoAssist Chatbot for MOSDAC

A smart AI-powered assistant designed to simplify user interaction with the MOSDAC Portal by enabling natural language queries about satellites, rainfall, cloud data, geospatial information, and more.

---

🚀 Project Overview

J.A.R.V.I.S (Just A Rather Very Intelligent System) is a chatbot and data retrieval system that:

• Understands natural language queries
• Retrieves relevant data or document snippets from the MOSDAC portal
• Presents results in a clean, conversational format with optional sources or visuals

---

🔍 Features (Completed)

✅ 1. Fully Functional Chatbot Interface

• Built with Flask and custom HTML/CSS/JS UI
• Supports user input + bot responses
• Displays thinking animation during long responses

✅ 2. NLP Engine

• Uses SentenceTransformers for vector embedding
• Switchable backends: Ollama's TinyLLaMA or other open-source LLMs
• Smart keyword matching, context understanding

✅ 3. Static Content Vector Store

• Scraper crawls all internal MOSDAC links & PDFs
• PDF text + webpage text cleaned and chunked
• Stored in FAISS vector DB with source tagging

✅ 4. Query Response Engine

• Embeds user query, runs similarity search
• Fetches best matching chunk from FAISS
• Displays answer + source
• Handles greeting ("Hi", "Hello") gracefully

---

⏳ Features (Planned/In Progress)

🔹 Smart Live Data Fetching

• Automatically retrieve weather/cloud/rainfall datasets from public APIs or NetCDF/CSV
• Convert numerical data into natural language

🔹 Knowledge Graph Integration

• Build relationship graph between satellites, sensors, regions, documents using NetworkX/Neo4j
• Visualize with PyVis or D3.js

🔹 Geo-Intelligence Support

• Support region-aware queries
• Use GeoPandas/Shapely to filter datasets by region
• Add optional Leaflet.js maps to responses

🔹 Automated Update Pipelines

• Pipeline 1: Weekly crawl of documentation, PDFs, FAQ
• Pipeline 2: Hourly/daily live dataset polling
• Use APScheduler or cron jobs

---

🚀 Tech Stack

Layer	Technologies Used
Frontend	HTML, CSS, JS, Flask Templates
Backend / API	Flask, Python
NLP / Chat Engine	Ollama (TinyLLaMA), SentenceTransformers
Vector DB	FAISS
Scraping / Extraction	requests, BeautifulSoup, PyMuPDF, Playwright
Geospatial (Planned)	GeoPandas, Shapely, Leaflet.js
Knowledge Graph (Planned)	NetworkX, Neo4j
Automation	cron, APScheduler (planned)

---

📅 Current Status

Module	Status
Chat UI + Response	✅ Done
Static Crawler + FAISS	✅ Done
PDF Extractor	✅ Done
Ollama LLM Integration	✅ Done
Smart Greeting/Typing UI	✅ Done
Live Data Polling	⏳ Pending
Knowledge Graph	⏳ Pending
Geo Processing	⏳ Pending
Feedback Learning	⏳ Pending

---

📊 Example Queries

• "Tell me about INSAT-3DR"
• "Where can I get cloud data for July 2022?"
• "What are the payloads of Megha-Tropiques?"
• "Download rainfall info for Gujarat"

Example.input.for.JARVIS.mp4

(reply from the ai bot may not be clear or up-to-date as the project is still in prototype stage)

---

🔧 Setup Instructions

# 1. Clone the repo
https://github.com/your-username/JARVIS-MOSDAC

# 2. Create virtual environment
python -m venv venv
source venv/bin/activate  # or venv\Scripts\activate on Windows

# 3. Install dependencies
pip install -r requirements.txt

# 4. Run Flask app
python app.py
You will see an URL (e.g. http://127.0.0.1:5000), run the URL on you local device.

Ensure mosdac_data.json and FAISS index (mosdac_index.faiss) are already built. If not, run the scraper:

python mosdac_scraper.py
python build_vector_store.py

---

💪 Team Credits

Made with passion for ISRO Hackathon 2025 by Team \Stark Agents.