hackathon
Neolectra
Point at a roof on a map and find out what solar and rainwater harvesting would actually pay back.
Every rooftop solar calculator I could find asks for your roof area in square metres. Almost nobody knows their roof area in square metres, so people guess, and the number that comes back is a guess wearing a decimal point.
So this one asks for an address instead. It pulls the imagery, extracts the roof outline with OpenCV, and works out usable area from that. You correct it if it is wrong, which is a much easier task than producing the number from nothing.
What the estimate is made of
Roof area is only the first input, and on its own it produces the useless answer: kilowatt-hours per year, a number nobody can act on.
The rest of the chain is what makes it decidable. Local irradiance and rainfall come from Open-Meteo for the actual coordinates rather than a national average. Panel layout is fitted to the extracted polygon, because a roof with a water tank and a stairwell on it does not hold the panels its area suggests. Consumption is priced against BESCOM’s slab tariff, which matters more than it sounds: Karnataka tariffs are stepped, so the units solar displaces are your most expensive ones, and a flat per-unit rate understates the saving badly.
The output is a payback period in years and a monthly bill after installation. Those are the two numbers someone actually decides on.
Rainwater, because the roof is already measured
The same polygon that sizes a solar array sizes a rainwater catchment, and the same rainfall series that feeds one feeds the other. Adding it cost almost nothing beyond a tank-sizing calculation, and in a city with the water situation Bangalore has, it is frequently the better answer of the two.
Where it stands
Built at HACKNXS at MIT and placed second. It is a hackathon build and it shows: the roof extraction handles clean rectangular rooftops well and struggles with the dense, irregular, partly shaded ones that make up most of the city, which is exactly the case that matters. It is not deployed, so there are no screenshots of it running here rather than staged ones.
The interesting piece to carry forward is not the computer vision. It is that pricing against a real stepped tariff changes the recommendation, and every consumer-facing calculator I looked at gets that part wrong.