Parag
coretracker.md
← All work
Personal toolAI-directedLocal-first

CoreTracker

A job-application tracker I built for myself, then automated with a scheduled AI task until it fills itself in.

CoreTracker v2 — the application board, sheet view, seeded demo data
Role
Design & build — self-initiated
Timeline
2025
Platform
Web · local-first
Status
Self-built · in daily use.

// the problem

Tracking a job search is its own job

A search across dozens of roles needs a system — what to apply to, what is in flight, what came back. So I built one for myself: a Kanban board (To Apply, Applied, Interviewing, Offer, Rejected, Ghosted) with a match score for each role against my résumé.

// the architecture

Local-first, no backend

The app has no server. Every record lives in the browser’s IndexedDB. My data stays on my machine; a visitor who opens the link sees an empty board. That is a deliberate choice for something holding real employers, real rejections, and a ghosting or two.

// the work

V1 to V3, and where the AI came in

  • V1 — manual: fill in a form to create a job card. No scoring of the listing against my résumé.
  • V2 — card creation streamlined, and a JD-vs-résumé match score added to every role.
  • V3 — automated: a scheduled Claude task finds the ten best listings matched to me, updates the tracker, and watches my mailbox to move a card the moment a reply lands.

V3 is the point. Every other project here describes an AI-directed workflow. This one is that workflow, shipped — the AI drafts, meaning it finds and files; I direct, meaning I decide what is worth applying to and how to pitch it.

The board — six columns with match scores (seeded demo data, not real applications)
The tracker at a glance. Match score sits on every card so triage is a scan, not a read.
V1 → V2 → V3 side by side — manual form to self-updating board
The arc that is the case study: a manual tracker that learned to maintain itself.

// outcome

A tool I actually use

CoreTracker runs my own search day to day. The honest measure is not a metric — it is that the thing works well enough to rely on, and that V3 turned a manual tracker into one that keeps itself current while I get on with the interviews.