SkinGraph
2025 · Researcher
A skincare tool that became a study in who gets left out of AI systems.
- Problem
- Vision models trained on narrow datasets quietly fail the people those datasets underrepresent — and the failure is invisible until you look for it.
- Outcome
- Reframed a product as a research question and surfaced concrete gaps in dataset coverage and model behavior across skin tones.
Why this project
SkinGraph started as a skincare tool and became a study in who gets left out of AI systems. The question it raised pulled me into computer vision research.
What I did
- Built the initial vision pipeline for the skincare use case.
- Probed model behavior across skin tones and surfaced where coverage thinned.
- Reframed the work from “ship a feature” to “measure who the system fails.”
Outcome
The project documented concrete gaps in dataset coverage and model behavior. More personally, it set the trajectory: from products, to computer vision, to the geography of how space and people are encoded in data.
Placeholder case study. Replace this prose, the cover image, and metadata with the real project.
Discussion