01Hackathon20262nd place, Hackwell 2026
Mind Armor
A browser extension that flags manipulative content and explains why
- My part
- Hackathon team
- Status
- Built, with parts designed or planned
- Stack
- React, WXT, Hugging Face
- Code
- GitHub ↗
Fig. — How data moves through itRead left to right
- 01Page you’re readingDOM text
- 02Ontology rulesnarrow to candidate claims
- 03Hugging Face modelsonly on candidates(model call)
- 04Highlight + reasoningpersonalised by interests
01Problem
The problem statement: use AI to detect manipulative content online. Around 30 to 40 teams picked it. Among the finalists, ours was the only team to build a browser extension, so it works on the page you are already reading.
02Approach
Ontology rules narrow down the page content first, so only the relevant parts go to Hugging Face models. Instead of stopping at the problem statement, I treated it as a product: a personal extension for everyday browsing, plus a school edition for media and AI literacy.
03What I built
- 01Onboarding that asks for your interests (sports, finance and so on) so results are personalised.
- 02Page scanning that highlights manipulative claims, like “grow hair in 5 days with this oil”. It doesn’t just say true or false: it gives logical and analytical reasoning so you stop and think before buying in a hurry.
- 03A teacher view for schools: the teacher shares a website, and sees each manipulative claim on it with its severity and a plain explanation.
04Not built yet
- designedStudent exercises: students get the same website with the answers hidden, find the manipulative content themselves, explain it, and are scored against the teacher’s answers.
- designedA subscription model for schools.