typesafe-adblock
🧹 Fun project: a Chrome extension that asks a tiny AI decision model (TypeSafe Jev) "is this DOM element an ad?" and pops it off the page. BYOK, no backend, not a real ad blocker.
🧹 Fun project: a Chrome extension that asks a tiny AI decision model (TypeSafe Jev) "is this DOM element an ad?" and pops it off the page. BYOK, no backend, not a real ad blocker.
The extension batches compact descriptions of candidate page elements into one request. Jev returns a Noul probability for each candidate asking whether it is a paid advertisement or sponsored placement; the extension applies its configured threshold before highlighting or removing elements.
Is page element candidates[0] a paid advertisement or sponsored promotional placement, judging its text, label, links, iframe source, attributes, and shape in page context?
A page hostname/title context and a representative compact element description, such as tag, label, text excerpt, link host, and element shape. The implementation batches candidates and numbers their question IDs.
Code extracts likely ad-shaped elements and excludes oversized or protected page content.
Create compact page and element metadata and cap the number of candidates per request.
Send one Noul question per candidate in a single request.
Highlight or remove elements whose returned probability meets the configured threshold.
Rescan after page mutations and scrolling to handle lazy-loaded candidates.
It confines the model to one semantic judgment per code-selected candidate. Candidate extraction, the operating threshold, and the DOM action remain explicit browser code.