01 / PERSONAL PROJECT

CareerOS

A Chrome side-panel extension that brings job details, fit evidence, AI advice and your own decisions into one workflow.

The problem

Job searching spreads information across vacancy pages, AI conversations and personal notes. A recommendation is less useful when its reasoning gets lost—or when it is mistaken for a decision already made.

I wanted a workflow that keeps the advert, the evidence and my own judgement connected.

The workflow

CareerOS opens beside a vacancy in Chrome. It reads the job details and produces a local fit snapshot, with separate areas for direct evidence, transferable evidence, likely gaps and manual checks.

For a second perspective, it prepares a prompt for the user’s chosen AI conversation. The user sends the prompt and brings the short response back into CareerOS. The AI recommendation and the user’s final decision remain separate.

01 / READ THE VACANCY

Keep the source in sight.

The side panel sits alongside the original advert, so extracted information can be checked and corrected before it informs a decision.

Ohpen vacancy beside CareerOS, showing the extracted company, Bratislava address, Hybrid work setup and 1 500–2 500 EUR per month gross. — Job details, in context. This example captures Ohpen, the advertised address, the gross monthly salary range and the occasional home-office arrangement as Hybrid. View full size ↗

02 / REVIEW THE EVIDENCE

A second opinion. A separate judgement.

The local snapshot and the external AI response are different steps. Here, ChatGPT recommends applying and supplies reasons, a main risk and a next step. CareerOS keeps the manual checks visible.

A ChatGPT Apply recommendation beside CareerOS’s local evidence snapshot, including a manual check for required years of experience. — Advice to review, not a hiring prediction. The AI output is a recommendation generated from supplied information; its claims still need checking. View full size ↗

03 / KEEP THE DECISION

Remember what you decided—and why.

History stores the user’s decision separately from the AI recommendation, alongside reasons and application progress. In the Support Administrator example below, the user chose Apply while the AI recommendation was Maybe.

CareerOS history showing two other vacancies, separate user and AI decisions, application stages and saved reasoning. — A separate history example. These are different vacancies from the Ohpen walkthrough. Decision, application stage and outcome are separate fields; the screenshot does not establish a hiring outcome. View full size ↗

The screenshot exposed the bug.

During portfolio preparation, the original reader missed the company and salary even though they were visible on the page. It also described occasional home office too broadly as “Remote or flexible.”

The reader was updated to include the facts panel above the description, recognise more salary formats and distinguish occasional home office from fully remote work.

The revised reader passed 26 browser-fixture tests. I then checked the vacancy again in Chrome: the first screenshot shows the corrected fields.

What I learned

A convincing AI response cannot make up for incomplete input. Checking the extracted facts against the source is a necessary part of the workflow, not an optional finishing touch.

Scope and limits

CareerOS is a personal prototype. Page layouts vary, extracted fields remain editable, and the fit snapshot is not a probability of being hired. This case study demonstrates the workflow and a concrete iteration; it does not claim measured job-search or hiring results.

These are screenshots of the working interface. The extension’s source code, private prompts and underlying profile documents are not included.

Next projectRecruiter Decision Copilot →