Ecowatt

Though this website might give the impression that I am a lone wolf, quietly building things by myself in a dark room, there is another side of me that genuinely likes to team up. Do not get me wrong, my loneliness is not killing me. But every so often it is good to work with people who have the skills you lack, and to be the person who has a skill they lack.

Ecowatt was one of those times. Five of us, from five wildly different fields, got dropped into the same brief with far too many opinions between us. The project started with military logistics. Then neighbourhood sharing. For a while in the middle it was probably three or four other things I have mercifully forgotten. Eventually we landed on something a French household actually loses sleep over: energy bills, and the mysterious letter grade (your DPE class) that quietly decides how much you pay to stay warm.

My job was the data, which is the part I would have volunteered for anyway. I got to quietly build a model that guesses a home's energy rating from a handful of inputs, feed it open government data, and feel briefly like a wizard, while everyone else did the harder work of making the thing make sense to actual humans.

Because then we showed it to real people. Around 21 of them, over two rounds of testing, who were completely unmoved by my accuracy score and instead asked reasonable things like "what does any of this mean" and "why does it cost that". The project slowly changed shape, from a clever data thing into something a person could actually use. Humbling, in the right way.

It is finished now, and we got a little ambitious with it: a free platform where people save energy while the data they share makes the tool smarter. Whether it becomes a real business or stays a very well documented school project is, pleasantly, still an open question.

Ecowatt helps French households make sense of their DPE energy rating and see what they can realistically do about it. It began as a Forward College project brief from Enedis on energy efficiency, and grew into a web tool built for two groups with the same blind spot: owners who can renovate, and renters who can only change habits. The aim was never to lecture anyone, just to show people what is actually worth swapping.

I owned the data side. That meant turning French open government data into a model that estimates a home's DPE label from a few accessible inputs, and setting up the pipeline (a Supabase backend) so the tool could keep learning from what people entered.

The model itself has its own write-up, so I will not rebuild it here: see DPE Label Prediction. Short version: it is accurate enough to be genuinely useful, and honest about where it is not.