CarboMeter: NLP Emissions Tracking
2023-12-01 · AI Engineering · Python · Django REST Framework · PostgreSQL · AWS · React Native · Expo
CarboMeter is a mobile app that lets an organisation record its greenhouse gas emissions by describing activity in ordinary language, instead of filling in a spreadsheet of fuel volumes and utility readings. It grew out of Carbonis, a group research project at SLIIT.
Why natural language
Emissions reporting fails at data entry. The people who know that a generator ran for six hours are rarely the people who know which emission factor applies to diesel, and asking them to learn is how reporting quietly stops happening. The app takes the description and does the classification and lookup behind it, so the reporting burden lands on the system rather than the person.
What it involved
Natural language processing and data preprocessing, model development in Python, and web scraping to assemble reference data. A Python and Django REST Framework backend on AWS with PostgreSQL alongside NoSQL storage, and a React Native mobile client built with Expo.
Recognition
Microsoft Imagine Cup 2023 World Finalist, and winner of the SLASSCOM National Ingenuity Award 2023.
Looking back
This was the project that pointed me at everything I do now. The interesting part was never the model; it was deciding what the system could safely infer from a vague sentence and what it had to ask about. That question turned up again in my dissertation and in the quotation service I built at work, and it is still the one I find most interesting.