DataLab is ITAM’s student data-science community. During 2025 I was its Research Manager and took part in its main project, ECOBOX AI: a computer-vision model that classifies waste into recycling categories from an image, meant for a smart bin. The final deliverable was the trained model and a paper; we did not reach a deployed prototype, and I would rather say so plainly.
The technical problem
Image classification with imbalanced classes and photos taken in real conditions: variable lighting, dirty or deformed objects, unhelpful backgrounds. The pipeline is the classic one of modern vision: a labelled dataset, data augmentation, a pretrained convolutional model fine-tuned to our classes, and an evaluation that looks at the full confusion matrix rather than accuracy alone, because confusing glass with plastic does not cost the same as confusing cardboard with paper.
The technical part was as expected. What I did not expect was everything else.
Doing research with students
A student team has a property no professional team has: everyone disappears during exam week, at the same time. Working on the project meant accepting that reality and designing around it.
- Planning in short milestones, aligned to the academic calendar, instead of an annual plan that breaks in October.
- Documentation as a product: every technical decision, every experiment and every dataset with its card, somewhere a new person could join and contribute in their first week. In a team with semester turnover, documentation is not an extra: it is the only thing that survives.
- Small, verifiable tasks, so someone with three free hours could close something complete.
- Meetings that end in decisions, and decisions written down.
What I take with me
That sustaining research as a team is a different skill from doing research, and the two need each other. That documentation is the infrastructure of continuity. And that working on a problem with environmental impact, even at the scale of one bin, gives the team a reason to come back after exams.