https://github.com/parthapray/air_pollution_llm_advisor
This repo shows the coding of air pollution LLM advisor chatbot for localized LLM using Ollama
https://github.com/parthapray/air_pollution_llm_advisor
airpollution chatbot edge india iot large-language-models localized ollama open-meteo-api privacy raspberrypi4 restapi
Last synced: 11 days ago
JSON representation
This repo shows the coding of air pollution LLM advisor chatbot for localized LLM using Ollama
- Host: GitHub
- URL: https://github.com/parthapray/air_pollution_llm_advisor
- Owner: ParthaPRay
- License: other
- Created: 2025-06-24T14:47:30.000Z (4 months ago)
- Default Branch: main
- Last Pushed: 2025-06-24T15:00:10.000Z (4 months ago)
- Last Synced: 2025-06-24T15:46:17.450Z (4 months ago)
- Topics: airpollution, chatbot, edge, india, iot, large-language-models, localized, ollama, open-meteo-api, privacy, raspberrypi4, restapi
- Language: Python
- Homepage:
- Size: 0 Bytes
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Air Pollution LLM Advisor
**Author**: Partha Pratim Ray
**Date**: 24 June 2025
**Email**: parthapratimray1986@gmail.com
---
## 📖 Project Overview
This project integrates:
- **Live air quality data** from the Open-Meteo API
- **Large Language Model (LLM)** reasoning using a locally hosted Ollama server
- **Interactive map** with Folium visualization
- **Conversational interface** using Gradio chatbot
- **Persistent CSV logging** of all data, including model metrics
The system provides both real-time pollution data and expert-like environmental health advice using a locally running LLM model. It allows users to:
- Search for any Indian city location
- Get air quality data visualized and summarized
- Query an LLM for health impact advice on the air quality
- Log all interactions (first LLM advice only) into a CSV file for research and future analysis
---
## ⚙ Deployment Scenario: Raspberry Pi + Local Network Ollama
This system is specifically designed to run on **resource-constrained edge devices like Raspberry Pi 4B**, while leveraging the power of **localized private LLM inference** hosted on a separate GPU-enabled machine on the same local network.
The Raspberry Pi 4B acts as:
* The **interactive user interface server** using Gradio (lightweight, fully browser-based UI)
* The **data fetcher** from Open-Meteo public API for live air pollution data
* The **controller** that sends structured prompts to the LLM server over the local network
Meanwhile, the Large Language Model (LLM) itself runs independently on a **local GPU-powered laptop or workstation** using **Ollama**, an open-source privacy-preserving LLM deployment platform.
This architecture offers:
* ✅ **Privacy-aware design** — no external cloud dependencies
* ✅ **Free of cost** — Ollama and all models run fully offline after setup
* ✅ **High availability** — even works entirely offline after initial setup
* ✅ **Separation of concerns** — lightweight front-end on Raspberry Pi, heavy model inference on GPU server
* ✅ **Interactive real-time dialogue** — Gradio handles persistent chat sessions, the LLM responds to queries, and the user can freely continue conversations after receiving advice
In this hybrid edge-GPU deployment, all sensitive data remains entirely within the user's private infrastructure, making it highly suitable for research, academic demonstrations, or privacy-critical deployments.
---
## 🌐 Features
- Live AQI data retrieval from Open-Meteo
- PM2.5, PM10, CO, NO2, SO2, O3, AOD, Dust, UV Index, Methane readings
- Folium-based India map with markers
- Gradio UI with full interactive chat
- Uses **Ollama LLM** (local deployment) to answer health queries
- First LLM interaction logged into `llm_aqi_log.csv` with full model metrics:
- `total_duration`, `load_duration`, `prompt_eval_count`, `prompt_eval_duration`, `eval_count`, `eval_duration`, `tokens_per_second`
- After initial LLM analysis, user can continue chatting but no further logs are generated (only first LLM call is logged)
- Fully edge-deployable with Raspberry Pi or local laptop server running Ollama
---
## 🔧 Requirements
- Python 3.10 or newer
- Ollama server installed and running locally (https://ollama.com/)
- Models loaded inside Ollama (e.g., `granite3.1-moe`, `deepseek-r1`, `qwen3`, etc.)
---
## 🔑 Installation
Clone this repository:
```bash
git clone https://github.com/yourusername/air-pollution-llm-advisor.git
cd air-pollution-llm-advisor
````
Create virtual environment (recommended):
```bash
python3 -m venv venv
source venv/bin/activate
```
Install Python dependencies:
```bash
pip install -r requirements.txt
```
---
## 🔑 Running Ollama Server
Install Ollama on your host machine:
```bash
https://ollama.com/download
```
Load your desired model(s) into Ollama:
```bash
ollama run granite3.1-moe
# or
ollama run deepseek-r1
```
Ensure Ollama API is available at:
```
http://localhost:11434/
```
If you're using a remote Ollama server, edit this line in the code:
```python
OLLAMA_IP = "192.168.x.x" # your Ollama host IP
OLLAMA_PORT = "11434"
```
---
## 🚀 Launching the Application
Start your Gradio UI by running:
```bash
python app.py
```
By default, it launches at:
```
http://0.0.0.0:7860
```
---
The Ollama activivity can be seen by following command on laptop acting as cloud in real-time on a terminal:
```bash
journalctl -u ollama -f
```
---
## 📝 Output File
The application creates a log file `llm_aqi_log.csv` containing:
* Timestamp
* Location & Coordinates
* All pollutant values (cleaned: no units)
* Chosen LLM model
* LLM generated response
* Ollama model metrics
Example:
| timestamp | location | lat | lon | PM2.5 | PM10 | ... | llm\_response | total\_duration | load\_duration | ... |
| --------- | -------- | --- | --- | ----- | ---- | --- | ------------- | --------------- | -------------- | --- |
---
## 📦 requirements.txt
```text
gradio
folium
requests
```
---
## ⚠️ Notes
* The first click on "**Ask LLM for Pollution Analysis**" triggers CSV logging.
* After that, multiple chats can continue but only the first advice gets logged.
* Units (μg/m³, ppm, mg/m³) are automatically stripped before logging.
* Compatible with any Ollama model that supports `/api/generate` or `/api/chat` endpoints.
* Thinking mode is disabled in Ollama calls (`"think": false`) for faster responses.
---
## 📄 Citation
If you use this work, please cite:
```bibtex
@misc{ray2025airpollution,
author = {Partha Pratim Ray},
title = {Air Pollution LLM Advisor},
year = {2025},
month = {June},
url = {https://github.com/yourusername/air-pollution-llm-advisor},
note = {Version 1.0},
howpublished = {GitHub repository},
}