Application Tracker

Local-first job application tracker. Paste a posting URL — company, title, location, and sponsorship signals are prefilled from the posting itself. Your data lives in a plain Excel workbook on your disk. No accounts, no cloud, no cost.

I built this in the middle of my own job search: a hundred-plus applications across two countries had outgrown my spreadsheet, but I didn't want to abandon the spreadsheet — so I grew a product around it. Everything here exists because I needed it that week.

Features

• One-field add — paste a job URL; the server resolves it through a parser chain: ATS JSON APIs (Greenhouse, Lever, Ashby, Workable, Recruitee, SmartRecruiters, Workday) → schema.org JSON-LD → OpenGraph/meta heuristics. Every prefilled value carries a provenance chip; nothing saves without your review.
• Blocked-site fallback — LinkedIn and Indeed refuse robots. The app detects the authwall and parses a pasted job description locally instead, including visa-sponsorship mentions with the matching quote as evidence.
• A pipeline that matches reality — four stages: Wishlist → Applied → Interview → Offer, with a closed tray for rejections. One-click advance / reject buttons on every card (with sound and a small celebration when an offer lands), drag-and-drop, and undo for everything.
• A hyperreal globe — applications pinned on a satellite Earth floating in a starfield (MapLibre globe projection + atmosphere over Esri World Imagery), entered with a cinematic space-to-Earth descent. Locations are geocoded once via OpenStreetMap Nominatim (cached, rate-limited, free) and stored as coordinates in the workbook; remote roles are listed, never pinned to fake spots, and unmapped rows get a one-click fix.
• Insights — response rate, Applied → Interview → Offer funnel (history-aware), weekly momentum vs. your goal, and source/location/sponsorship breakdowns. Computed from your data; nothing is estimated.
• Interview prep bank — questions and answers, grouped by category.
• Excel is the database — open tracker.xlsx in Excel any time. External edits are detected and reloaded; if the file is open in Excel during a save you get a clear banner instead of a corrupt file. Atomic writes plus rolling backups in data/backups/.
• Command palette (⌘K), keyboard-first navigation, dark/light themes, mute toggle.

Pipeline	Map

Add flow	Insights

Quick start

Requires Python ≥ 3.10.

The easy way — clone (or download) the repo, then:

• macOS: double-click start.command (first time: right-click → Open if macOS asks)
• Windows: double-click start.bat

The first launch builds its own environment in about a minute; every launch after that just opens the app in your browser. If it's already running, the launcher simply brings it up again.

The terminal way:

git clone https://github.com/jamie-dongjae/application-tracker && cd application-tracker
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
python run.py

Either way the server binds to 127.0.0.1 only — nothing is exposed to the network.

You don't need to bring any Excel file. On first run the app creates data/tracker.xlsx with the right sheets and columns, and every change you make is saved into it. It stays an ordinary workbook you can open, sort, and filter in Excel. Optional: python scripts/make_sample_data.py fills it with 30 fictional rows to explore first.

Already tracking in a spreadsheet?

python scripts/import_legacy.py --in "/path/to/Old_Tracker.xlsx"

The importer finds your header row by column-name aliases (Company, Job Title, Status, Location, URL…), coerces dates, canonicalizes statuses, and never modifies your source file. Then run Geocode unmapped locations from the command palette to pin imported rows on the map.

Your data vs. the demo

Everything under data/ (your workbook, caches, backups) is gitignored and never leaves your machine — CI fails the build if a workbook ever lands in the repo. The live demo is a fully static build with fictional sample data; the real app is what you run locally.

Architecture

flowchart LR
    B[Browser<br>vanilla JS, no build step] -->|JSON| A[FastAPI · 127.0.0.1]
    A --> S[ExcelStore<br>openpyxl · atomic writes · backups]
    S --> X[(tracker.xlsx)]
    A --> P[Parser chain<br>ATS APIs → JSON-LD → meta → pasted text]
    A --> G[Nominatim geocoder<br>cache · 1 req/1.1 s]

Loading

Five runtime dependencies: fastapi, uvicorn, openpyxl, httpx, selectolax. The frontend is plain ES modules — no bundler, no node_modules. Sounds are synthesized with WebAudio (no audio files); motion (staggered entrances, KPI count-ups, kanban Flip glides) by GSAP — the app degrades gracefully if its CDN is unreachable and respects prefers-reduced-motion. Satellite imagery © Esri, Maxar, Earthstar Geographics; labels by CARTO/OpenStreetMap; rendering by MapLibre GL.

Development

pip install -e ".[dev]"
pytest

All tests run offline against committed fixtures (real ATS payload shapes, JSON-LD pages, authwall pages, pasted job descriptions). CI runs on Python 3.10 and 3.12. Older workbooks (v2 schema: five-stage pipeline, salary columns, STAR prep) upgrade in place automatically on first load.

License

MIT