Question
Ask in plain language“Which city had the highest average PM2.5 in 2023?”
Local Gemma
Plans a compact DSL[rank · city · PM2.5 · 2023]
Evidence
CPCB measurementsIngest and normalize once
Guardrail
Validate + execute locallyCheck city · pollutant · dates
Checked output
Checked resultAnswer · table · plot
The model plans. Code verifies and computes against packaged ground measurements.
What's new · October 2026
Checked answers over official CPCB data, weather on a grid from our own WRF model, offline phone apps, and where VayuChat is being used, with CEEW, in Thane and New Delhi.
New · Weather on a grid
VayuChat can now answer questions over gridded weather from the Sustainability Lab's own WRF simulation, not only over monitoring stations: a value for every 0.25° cell of India, on every day of 2023 and 2024. Ask about a heatwave or a cyclone and get back a map or an animation you can download.
“Map the mean maximum temperature across India during the 25–31 May 2024 heatwave.”
The answer is a map of the week's mean daily maximum, up to 48.6 °C across the north, with the states and cities that were hottest. Ask for an animation instead and VayuChat plays it day by day.
“Animate daily rainfall across Rajasthan from 13 to 21 June 2023 as Biparjoy's remnants crossed it.”
This animation is VayuChat's own GIF export, exactly as downloaded. Every map has Download image and every animation Download GIF, ready for a report or a slide.
WRF v4.7.1, domain d01 (27 km), run by the Sustainability Lab, IIT Gandhinagar. These are model simulations, not observations. Weather questions are in testing on the lab's VayuChat build and coming to the hosted app.
Three interfaces
Use the web workspace for deeper, source-linked analysis. Use either phone app when questions and data must stay on the device.
Compare cities, inspect coverage, make maps, and trace every result back to its source and analysis recipe.
Current React interface · four-city comparison
Airplane-mode answers, tables, and Swift Charts from packaged Indian air-quality data.
Direct answers, dataframes, plots, and an inspectable plan—without a network connection.
In the field · with CEEW
VayuChat is built with the Council on Energy, Environment and Water (CEEW), which has taken it from the lab to the people who use air-quality data every day.
Thane Municipal Corporation's Air Quality Decision Support System, built with CEEW, uses VayuChat to answer questions in Marathi and English over data from 350+ construction-site sensors. The system is featured in CPCB's Compendium of Best Practices under the National Clean Air Programme (September 2026).
At CEEW's Clean Air Horizons 2026, about 30 researchers, policy analysts and students tried VayuChat with their own questions. Photo: CEEW.
CEEW's Arunabha Ghosh and Dr Mohammad Rafiuddin cite VayuChat as AI built around the people who use it.
Read the op-edEarlier: VayuBuddy, VayuChat's first version, was covered by the Indian Express, Ahmedabad Mirror, Divya Bhaskar and Chitralekha in December 2024.
Research
VayuBench evaluates whether a model can produce an executable program and the correct answer—not merely fluent text.
Primary paper · CODS 2025
The benchmark contains 5,000 natural-language queries paired with verified programs across seven query families. The study evaluates 13 open models under a common schema-aware protocol and follows the work through to VayuChat.
@inproceedings{acharya2025vayubench,
title={VayuBench and VayuChat: Executable Benchmarking and Deployment of LLMs for Multi-Dataset Air Quality Analytics},
author={Acharya, Vedant and Pisharodi, Abhay and Pasi, Ratnesh and Mondal, Rishabh and Batra, Nipun},
booktitle={13th ACM IKDD International Conference on Data Science},
year={2025},
doi={10.1145/3799830.3799884}
}Demo paper · CODS 2025
VayuChat: An LLM-Powered Conversational Interface for Air Quality Data Analytics