At 7:14 on a Tuesday morning in November, a sensor in our neighbourhood reads AQI 412. The number is a fact. A neighbour comes back from a walk short of breath and types into the JanVayu app: “the air today tastes metallic, my chest is heavy, I won’t take the children out”. That is also a fact.
Neither fact by itself tells the school principal four streets away whether to keep the children indoors for assembly.
JanVayu has spent two years collecting both kinds of fact side by side. The sensor readings are dense and continuous. The testimonies arrive in bursts, in vernacular, with detail no instrument records: what the air tastes like, whose breathing is laboured, which window stays shut. Building the repository was the easy part. The hard part was deciding what to do with it.
What the assistant does
A sensor on the left, the AI in the middle, a resident testimony on the right. The two streams feed in and the assistant sits between them.
The assistant takes a question from a resident (“should we hold assembly outdoors at 7:30am?”, “is it safe for my asthmatic mother to walk to the temple this evening?”, “what is the air doing on the lane behind the school?”) and draws on both streams at once. It pulls the sensor readings for that hour and that lane, and the testimonies people have left in the same window. Then it gives an answer the resident can act on, for example: “the readings are high and rising, residents nearby report the air as heavy, defer assembly to 8:30 or move indoors”.
That is all it does. It does this one job and declines every other.
What the design refuses
Most AI tools for environmental data are built on the opposite bet. They promise prediction, dashboards, and summaries of a city’s air quality at country scale. Builders are tempted to serve the policy audience because that audience pays.
The JanVayu assistant does none of this. It does not predict tomorrow’s AQI, because existing models already do that and a resident is rarely asking for a prediction. It does not produce a city-level summary, because a city average hides the neighbourhoods that are worst off. It does not rank neighbourhoods against each other, which would be a different and more dangerous project.
It sits between one number and one testimony and translates between them, for one person at one moment. The technology is only as ambitious as that.
What this is an argument about
Air quality measurement has long treated its two sources unequally. The sensor counts as the fact and the resident counts as the experience. Reports talk about reconciling the two and then mostly describe the sensor, with an anecdote added for colour.
On a phone, the assistant looks like this: a short question goes up and a longer, grounded answer comes back.
The assistant gives the two sources equal standing as inputs. The reading is precise about parts per million. The testimony is precise about what the air tastes like and who is struggling to breathe. Together they produce something neither produces alone, which is advice a particular person can act on, on a particular morning, for particular children.
This is a substantial claim. Most evaluation practice in this country still treats the sensor as ground truth and lived experience as confounding noise. JanVayu was built on the opposite assumption, and with the AI assistant that assumption becomes a working tool as well as a stated value.
What is next
The assistant is live and still learning. We are watching the cases where the two streams diverge: a morning when the sensor says 220 and the neighbourhood reports heavier air, or an afternoon when the sensor says 90 and residents in three lanes report eye irritation. We treat these as findings and not as errors. Each divergence shows one source picking up something the other misses, and points to where public-interest air measurement needs to pay closer attention.
If you see the assistant being asked the wrong question, or a question it should decline, please write in. Each time somebody pushes back, the tool gets better.