Project Neem: How Hot Does a Schoolyard Feel to a Child? Physics, an AI Planner and 30 Years of Shade
Neem shows how hot every square metre of a Delhi schoolyard feels to a child on a heatwave day, plans the trees that would cool it, and checks every plan with real radiation physics. It covers 766 schools, runs on AWS open source from a single laptop, and grows the shade thirty years ahead, so schools, scientists and funders can see what a tree is worth before it is planted.
The air said 42 °C. The schoolyard said 47.
It is half past twelve on a Monday in late April, and the bell has just gone at a government school in Rohini, in north-west Delhi. The morning shift spills out of the classrooms and into the open yard. There is no shade to walk through. The sun is almost directly overhead, and the ground has been soaking up heat since seven in the morning. It gives that heat back now, up through thin school shoes and into the faces of children who are nine years old.
The weather report for that hour said 42 °C. That is not what those children felt. They felt 47.
By the time they reach the gate, a child of nine has lost nearly a litre of water since morning assembly. Nobody planned it that way. Nobody chose it. It is simply what an open yard does to a small body on a Delhi summer day, and it happens in hundreds of schools at once, every summer, while the rest of us read "42 °C" on our phones and think we understand the day.
This is a story about that gap between the number we see and the heat a child lives in, and about what it would take to close it.
Why I care
I grew up in Cuttack, a small city in Odisha, where on a good day the air quality index drops to 20 or 30. When I came to Delhi to study at Amrita, my phone started showing numbers I had never seen at home. It wasn't panic. It was a quiet worry that would not leave. I kept thinking about the children growing up here, and about the decades all of us are walking into.
I couldn't fix the air on my own. But I kept coming back to something smaller and more hopeful. Shade is something a school can actually change. A tree planted this year changes the ground beneath it for the rest of our lives.
The question nobody could answer
Every principal already knows trees help. What none of them can know is where to plant, how many trees it takes, and what one tree will actually do for the children standing under it. And a neem takes about five years to grow a proper crown. Plant in the wrong place, and you find out half a decade later.
So I built Neem, named after the tree that grows in hard, dry ground and stays in full leaf through the hottest months. Neem answers that question before anyone picks up a spade.
Looking at a whole city
First, I wanted to see Delhi the way a child in a schoolyard sees it. Neem looked at the open ground of 766 schools across the city, all on that same hot Monday, all under the same sky.
The pattern was almost too clear. One yard in three spends more than a quarter of the school morning in extreme heat. The leafiest yards run nearly 3 °C cooler than the barest ones. And across all 766, the amount of tree cover and the heat in the yard move together so closely (a correlation of −0.95) that shade explains almost everything. Every yard had the same weather. What made the difference was what had been planted there, or not.
Why a thermometer doesn't tell the truth
A thermometer measures the air in the shade. But a child in an open yard isn't in the shade. They are standing in a field of heat coming at them from every direction: sunlight from above, glare from the sky, light bouncing up off the ground, and warmth radiating off every wall and paving stone that has been baking since sunrise.
At the centre of that yard in Rohini at 1 pm, the air was 42 °C, but all of that radiation together hit a body like an oven at 67 °C. Add in the hot, dry wind and the result is a feels-like temperature of 47 °C. The line for extreme heat stress is 46.
To calculate all this, Neem uses SOLWEIG, a model built by climate researchers at the University of Gothenburg. For every square metre of a school, it works out where the sun falls and where buildings and trees cast shade, how much open sky each spot can see, and how much heat the ground and walls throw back. On my laptop, that takes about four seconds per school.
The information it needs is all open, and most of it is stored on AWS. Building outlines come from Overture Maps and tree heights from a global map published by Meta and the World Resources Institute. I made one rule for myself: never download more than I need. Delhi touches only 3 of the 512 files in the building data, so Neem reads just those three. That's 1.9 million buildings in about two minutes, for the cost of a few clicks.
Making it feel instant
Four seconds is fast for science and slow for a person. If every tree you planted on the screen took four seconds to show its effect, you would stop playing with it, and the whole point is to play, to try, to wonder "what if we put one here?"
So Neem does something simple. It doesn't recalculate the whole school when you plant a tree. It only works out what changed: which squares just fell into the new tree's shade, and how much cooler shaded ground gets, learned from that same school's own full calculations. Every square you didn't touch keeps its exact answer. The result appears in about a hundredth of a second.
I tested this shortcut against the full calculation on a layout of trees it had never seen, and across the whole yard it stayed within 1 °C. Right under a brand-new tree at midday it can be off by a few degrees, so the page tells you that honestly, instead of pretending to be perfect.
A child, not an average
Feels-like temperature is calculated for a standard adult. But the people in these yards are nine.
So Neem runs a model of the human body called JOS-3, which simulates how a body moves heat around and sweats, minute by minute. I shrank it to the size of a nine-year-old and walked it through a real school morning: assembly in the yard, class, half an hour of running around at recess, more class, and the walk home. Recess alone pushes the child's core temperature up by nearly a degree. By a quarter to one, they have lost 877 millilitres of water.
The body model was originally built for adults, so I only report what holds up for a child: how much their temperature rises from their own resting point, and how much water they lose.
An AI that isn't allowed to guess
Nobody should have to place 85 trees by hand, so Neem has an AI planner. It runs on a small open-source language model, right on my laptop, built with AWS's Strands Agents framework.
Small models can be careless. Early on, this one once answered without doing any of the work, and once spent past its budget. So I changed the rules instead of pleading with it. The AI is allowed to decide, but it is never allowed to calculate. It works only through tools. The tools place each tree on the hottest open ground. The tools refuse to spend a rupee over the budget. And before anything is shown to a person, the tools run the full physics calculation again. If the AI forgets to check its work, the code checks it anyway.
For Kendriya Vidyalaya, with a budget of three lakh rupees, it planned 85 neem trees for ₹2,97,500. The physics confirmed it: at 1 pm, the yard drops from 46.6 °C to 44.8 °C, and the child's body heats up less. Later I reran every plan from scratch, and every number came back exactly the same.
Thirty years, in one scroll
A sapling does nothing in its first summer, so judging a plan on day one is unfair to the tree. So Neem grows them.
It takes the same 85 trees and recalculates the same hot Monday at 1, 5, 10, 20 and 30 years. The weather never changes and the buildings never change, so the only thing different from one run to the next is the trees. The first year, nothing happens; the saplings are just two metres tall. By year five, the shade has arrived and the yard is out of extreme heat. By year thirty, it is a grove: 85 trees whose crowns cover nearly 6,000 square metres, and the ground beneath them feels like 42 °C, where today it feels like 47.
On the website you can scroll through those thirty years and watch the school grow green in 3D. How big a neem gets by each age is my estimate, based on published ranges for the species, and I assume every tree survives. That is a hopeful assumption, and I say so.
Built with AWS open source
I built Neem for the WeMakeDevs × AWS Environmental Hacks, in the Heat & Water track. I chose the Build It route: everything built with AWS's open-source tools and run entirely on my own laptop, with no cloud account and no bill.
Strands Agents powers the AI planner and its tools. AWS SAM describes the whole system as it would run in the cloud: storage for the data, a queue for physics jobs, workers that run the calculations, and a way to rank every school in parallel. LocalStack pretends to be the cloud on my laptop so all of that can be tested for real, and it caught a genuine bug that my ordinary tests had missed. Cedar writes down who is allowed to do what. Anyone can explore. Nobody, not even an administrator, can approve a planting plan that the physics hasn't verified.
I'll be honest about what isn't done yet. Nothing runs in the real cloud so far, and the part that ranks every school in parallel is written but hasn't been run; for now the ranking ran on my laptop. Open data also knows the heights of only about one Delhi building in two hundred, so most heights are estimates. Those are the next things to fix, and the code is ready for them.
Why start with schools
Children feel the heat first and longest. Every neighbourhood has a school, with open ground and a gate that opens every morning. And a school has one thing no street or park has: the people watering the saplings are children. The child who waters a tree at nine might plant a whole street at thirty.
Where this goes
Next, a teacher could walk an empty schoolyard with a phone for one minute, and an AI vision model would turn that video into real building heights. That code is written; the first real yard is next. The same video could become a 3D model of the school you can walk through on a screen, using World Labs' open-source Spark, with tomorrow's shade painted on today's ground. Measured again each summer, it could show which saplings survived and what they are doing for the children beneath them. That is the kind of honest report every NGO and foundation that plants trees has always wanted.
Nothing about this is limited to Delhi. The data covers the whole planet. India alone has nearly fifteen lakh schools. And beyond schools there are bus stops, hospital courtyards, streets and parks, every place where people wait in the sun.
The real change
The change I care about most isn't in the software. It's in the children.
A child who plants a tree at nine and stands in its shade at nineteen will understand, in their own body, what nature does. Not from a textbook. From the cool ground under their feet on a day that should have been unbearable. I believe a child who feels that will love trees for the rest of their life.
If Neem can help a generation fall in love with trees, and prove with physics that they're right to, then a city can slowly learn to grow its own shade. Neem where the ground is hard and dry. Banyan and peepal where there is room to spread. And one day, that particular hush you only find under very big trees.
The air said 42 °C. The schoolyard said 47. With 85 trees and a little patience, it can say 42 again.
Built by Ritesh Roshan for the WeMakeDevs × AWS Environmental Hacks, Heat & Water track.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article