Red-Ink

Inspiration

Children encounter speeches, interviews, debates, and other media online, where misleading claims can be hard to recognize in the moment. RedInk was built to help parents and educators notice claims that may need a closer look and use them as opportunities to talk with children about reliable information.

What it does

RedInk is a Chrome extension that listens to audio from a browser tab, transcribes it with Deepgram, and asks a language model hosted by Featherless to assess claims in context. It displays each claim, speaker label, assessment, and short explanation in a newspaper-style sidebar, so an adult can review it alongside the media.

How we built it

We built RedInk as a Manifest V3 Chrome extension using vanilla JavaScript. An offscreen document captures tab audio and streams it to Deepgram. The service worker batches transcript segments and sends them to a Featherless-hosted model, then saves structured findings in Chrome storage. A content script displays the results in a sidebar isolated with Shadow DOM.

What we learned

Building RedInk involved coordinating browser audio capture, live transcription, and model responses. We learned to buffer transcript fragments into more complete thoughts and request structured results that are easier to review. We also designed the output to be brief, so it can support a conversation without taking over the viewing experience.

Child safety and limitations

RedInk is intended as a support tool for adults guiding children through online media. It does not monitor a child, block content, or guarantee that a claim is true or false. Its

Built With

• chrome
• chrome-extension-api
• chrome-storage-api
• deepgram
• featherless-ai
• html
• javascript
• manifest-v3
• mediarecorder
• qwen
• shadow-dom
• web-audio-api
• websockets