A job search that can explain itself.

I turned my own job search into a design problem: how do you keep research, documents, applications, and responses organized without losing what each record actually proves?

MY CONTRIBUTION

Problem framing, information model, workflow design, evaluation, and iteration. Built with AI coding assistance.

THE RESULT

An operating local dashboard and coordinator, with public-board research, reviewed document preparation, and evidence-backed application records.

FICTIONAL WORKFLOW / NORTHSTAR LABS

What does “done” actually mean?

An illustration of the evidence model. These invented records do not submit an application or connect to a live system.

REVIEWED · NOT SUBMITTED

The file is ready. The application is not complete.

INPUT / OBSERVATION

A fictional résumé passes its claim and layout review.

RECORDED RESULT

Package: reviewed Attempt: none Employer outcome: unknown

A document review proves readiness. It provides no evidence of an employer receiving the application.

All example outcomes:

Package ready: REVIEWED · NOT SUBMITTED

A fictional résumé passes its claim and layout review.

Package: reviewed Attempt: none Employer outcome: unknown

A document review proves readiness. It provides no evidence of an employer receiving the application.

Interrupted: UNCERTAIN · RECONCILE

A worker reserves an attempt, starts transmission, then loses the result.

Package: reviewed Attempt: uncertain Next action: inspect employer evidence

Another worker must not infer success or blindly submit again. The uncertain attempt remains visible until reconciled.

Confirmed: RECEIPT OBSERVED

The fictional employer displays a submission confirmation.

Package: reviewed Attempt: confirmed Employer outcome: no hiring decision yet

A submission receipt proves receipt of an application. It is separate from a recruiter response or interview invitation.

THE PROBLEM

What needed to change.

A promising role, a finished résumé, and a submitted application are three different things. When those states blur together, a dashboard can look productive while hiding unfinished work. The first version needed a way to reconcile records without turning assumptions into facts.

THE DECISIONS

Give each kind of evidence its own place.

The original Control Room keeps discovery, document readiness, application attempts, and employer outcomes separate. A removed posting does not become a rejection. A prepared file does not become a submission.

Make interrupted work recoverable.

V2 adds persistent work ownership and an explicit attempt history. A worker reserves an attempt before transmission. If the result is uncertain, the next action is reconciliation rather than an automatic second submission.

Review the exact file being used.

Candidate facts stay separate from generated prose. Packages are checked for supported claims and rendered layout; file hashes connect the reviewed document to the later attempt.

CURRENT STATE

What works today.

The local implementation has completed fictional recovery scenarios and requested research against public employer boards. It has also recorded an agent-assisted application with actual employer confirmation. The agent operated the form; the coordinator retained the evidence.

Unattended form execution, recurring scheduling, and mailbox monitoring are still unfinished. This public demonstration is fictional and has no connection to the private job-search system.