Last update: Aug 21th
Tired of scrolling through hundreds of LinkedIn job postings, reading long descriptions, and facing misleading experience requirements, I built WhatYOE — a Safari web extension powered by Apple’s on-device Foundation Models.
WhatYOE analyzes job descriptions in real time, highlighting candidacy directly on each job card to provide immediate, private feedback on whether the user qualifies.
WhatYOE also saves each analysis to a local desktop interface, enabling users to organize both evaluated jobs and resumes. The system is designed for fully automated analysis, streamlining the entire screening and organization process.
WhatYOE is not meant to be a smart-magnifier to help you pick apples one by one
WhatYOE is an automated, hands-free apple picking pipeline.
Illustrations generated by ChatGPT
Journey map
&
Workflow
Last update: Aug 17th
: Implemented
: Pending
First Time Setup
User imports resume through the Desktop App, which the Backend processes with AI for optimal matching. The cleaned resume is stored with a unique ID, ready for job analysis.
Job Discovery
User activates the WhatYOE Extension in Safari, selects their resume, and the extension automatically scans LinkedIn job listings. The Backend analyzes each job description against the selected resume, categorizing opportunities by match quality.
Review and Action
User opens the Desktop App to browse jobs organized by match rating and filtered by resume. They can quickly identify promising opportunities, jump directly to LinkedIn listings, and "like" jobs to prevent auto-deletion after 30 days.
Adaptive algorithm
Update Aug 21
Try it yourself
The new algorithm extracts experience, education, skills, required YOE, and actual YOE, then dynamically balances the weight of experience vs. education based on how much experience is required.
It ensures education matters more for junior roles, while experience dominates for senior roles.
How it works
(Algorithm outdated)
When a user imports a resume, the system cleans and structures the content, then generates and stores an LLM-distilled version for future evaluations. During analysis, the job description is also distilled to extract only the most relevant requirements.
The distilled data is then passed through four independent evaluation phases—Years of Experience, Education, Technical Skills, and Relevant Experience—each designed to focus on a single criterion and minimize long-context bias. Every phase produces two outputs: a Fit Score (how well the candidate matches) and a Gap Score (where they fall short). These scores are weighted using fit and gap multipliers to account for LLM nuances, resulting in a balanced and reliable final compatibility score.
System Component
WhatYOE is a dual-interface resume–job matching tool with a separated architecture: a Safari web extension for fast, real-time analysis and a macOS desktop app for detailed review and management. Both share the same on-device AI engine but work independently, enabling optimized performance, offline processing, and complete data privacy.
Progress Updates
I'm happy WhatYOE is close to MVP stage, and it's my second project own end to end development.
However, the progress will be paused for a while due to personal matters (deal with the consequence of unemployment).
WhatYOE scans a job in 30 sec while I can be rejected in 10 sec.
Owning a farming tool seems won't change anything.