Snitch
A second look at manipulative screens.
Snitch grew out of a personal build around the September 2026 Build with AI bootcamp in Bengaluru. It explores a simple question: can a screenshot help someone notice a misleading choice before acting on it? The app uses a vision model to surface possible dark patterns, explain the visible evidence and reference India's 2023 CCPA guidelines. Findings are reviewable, grievance drafts are editable, and sharing is user-triggered. It is an assistive prototype, not legal advice or a verified legal verdict.
- MY ROLE
- Developer - personal / event build
- PROJECT STATUS
- Public prototype
- BUILT WITH
- React / TypeScript / Vite / Gemini API / Vercel
A problem
worth exploring.
A confusing checkout or subscription screen can make an unwanted action look like the obvious next step. Recognizing a possible dark pattern is only useful if someone can inspect the evidence and decide what to do about it.
An end-to-end prototype for moving from a screenshot to reviewable findings and a draft grievance, while keeping the final interpretation and next action with the user.
How I approached it
- Send screenshots to a server-side vision-model integration and structure the response into findings with evidence and confidence.
- Connect findings to locations on the screenshot and references to the dark-pattern guidelines so the reasoning can be reviewed.
- Offer an editable grievance draft and an optional share card, leaving filing and sharing decisions with the user.
What I took away
- For AI-assisted review, showing evidence and uncertainty matters as much as producing an answer.
- Drafting a complaint and submitting one are different responsibilities; the interface should keep that boundary explicit.