ChillLens

Facebook simulation picture
Chat room simulation picture

Inspiration

Chill Lens helps people recognize manipulative language before they react or share.

Social posts and private messages often use urgency, fear, or social pressure to influence readers. We wanted to make these tactics visible while protecting the privacy of sensitive conversations.

What it does

Chill Lens highlights suspicious sentences, explains the persuasion tactic, and offers a short reflection question.

It detects five tactics: false consensus, manufactured urgency, fear of missing out, fake authority, and emotional polarization.

For example, “Everyone is sharing this—forward it now!” may trigger consensus and urgency cues. Chill Lens then asks: “What evidence supports this claim?”

The goal is to encourage reflection, not automatically decide whether a message is true or false.

How we built it

Local inference protects sensitive text. Because private chats may contain personal information, we run the Laya decision model on the user’s device instead of sending messages to a cloud model.

Our prototype includes a Facebook/Messenger simulator and a Chrome extension for Facebook feeds. Both share the same detection logic. Demonstration cases come from Taiwan’s Cofacts dataset.

Challenges we ran into

Facebook’s changing page structure makes reliable highlighting difficult. Traditional Chinese predictions can also produce false alarms.

To reduce unnecessary alerts, the system evaluates the overall post before identifying sentence-level tactics.

Accomplishments that we're proud of

We built an interactive simulator, a browser extension, and sentence-level explanations with reflection prompts.

Shared detection logic keeps both environments consistent, while ZIP export makes the extension easy to download and test.

What we learned

Helpful alerts explain what triggered them and leave the final decision to the reader.

Privacy also needs to be part of the architecture: sensitive text stays on the device during model inference.

What's next for Chill Lens

Improve Traditional Chinese and English detection, optimize browser-native local inference, and expand live support to more social and messaging platforms, including LINE Web and WhatsApp Web.