Job Pipeline

A Next.js app that finds remote developer jobs every day, filters them by country, scores them against your three CVs, logs everything in a Google Sheet, and gives you a dashboard to approve and apply from. Built for Daniel Chimezie. Runs on your PC for testing and on a RackNerd VPS for the daily run.

cron (7am daily)  ->  POST /api/run
                         |
   1. lib/fetchers/      pull jobs from 9 sources
   2. lib/filters.js     keep eligible countries, drop duplicates
   3. lib/scorer.js      score each job, pick the right CV
   4. lib/sheets.js      write the best new matches to the sheet
   5. lib/emailApply.js  find application emails in descriptions
   6. approvals          send email applications, queue the rest
   7. lib/mailer.js      follow-up reminders and one daily digest

Dashboard (app/page.js) reads and updates the same sheet.

1. Open the dashboard. Jobs are sorted by match score.
2. Approve a job you want. For an email job, your CV is sent at once (after a confirmation that shows the address). For a web-form job, it moves to Needs Manual Apply.
3. Open Apply kit on a job to copy your details and saved answers and to download the right CV. Open the application, fill it in, submit.
4. Click Mark Applied. A follow-up reminder is set for 7 days later.

Jobs scoring at or above AUTO_APPROVE_MIN_SCORE skip step 2. Only email jobs are ever submitted for you; web forms are always submitted by you.

resumes/ holds your 3 CV PDFs. cron/ holds the scripts that call /api/run daily. logs/ holds the daily run log.

pnpm install
cp .env.local.example .env.local     # then fill it in
pnpm dev                             # open http://localhost:3000

Google setup: create a Google Cloud project, enable the Sheets and Drive APIs, make a service account, download its JSON key into credentials/, create the sheet, and share it with the service account's email as Editor. Then run once:

curl -X POST http://localhost:3000/api/format
curl -X POST http://localhost:3000/api/run -H "x-cron-secret: YOUR_CRON_SECRET"

(If you set a dashboard login, add -u USER:PASSWORD to the format call.)

Keep the project in a plain folder such as C:\Users\YOU\job-pipeline. Do not put it in OneDrive: syncing breaks node_modules.

node scripts/check-companies.mjs                 # report
node scripts/check-companies.mjs --fix           # remove dead names
node scripts/check-companies.mjs --add=greenhouse:dropbox,ashby:retool

Restart pnpm dev after changing the list, because the fetchers read it at startup.

1. Install Node 20 and pnpm, upload the project, run pnpm install and pnpm build.
2. Fill in .env.local with Linux paths. Set ALLOW_CLEAR=false and set a dashboard login.
3. Start it with PM2: pm2 start "pnpm start" --name job-pipeline, then pm2 save and pm2 startup.
4. Fix Windows line endings in the scripts, then install the daily job: sed -i 's/\r$//' cron/*.sh && chmod +x cron/*.sh && ./cron/setup_cron.sh.
5. Block everything except SSH with the firewall and reach the dashboard through an SSH tunnel: ssh -L 3000:localhost:3000 user@your-vps-ip.

• .env.local and credentials/ must never be committed. Check with git check-ignore -v .env.local credentials/keys.json.
• The dashboard login uses Basic Auth, which is readable on plain http://. Use the SSH tunnel or HTTPS.
• config/applyKit.json holds your saved answers. Do not commit a real salary number to a public repo.
• Rotate the service account key, app password, and CRON_SECRET if they are ever pasted somewhere public.

• Only email applications are sent for you. Web forms (most jobs) are submitted by you; the Apply kit makes that faster.
• Scoring is a keyword-based first pass. Read the posting before applying.
• Country matching is text based and can miss unusual wording. Extend SYNONYMS in lib/filters.js when it does.
• Salaries are not read from postings, so the sheet shows "Not listed".
• Email detection can pick the wrong address. Check the address on the dashboard card before approving.
• LinkedIn and Indeed are not sources: neither offers individual API access.

1. Watch a few real runs and tune the word lists in lib/scorer.js and MINIMUM_MATCH_SCORE.
2. Run check-companies.mjs every month and add companies from your target list.
3. Deploy to the VPS (section above).
4. Build the LinkedIn alert-email reader.
5. Build an assisted-apply script if submitting forms takes too long.
6. More sources: NaukriGulf for UAE and Qatar, StepStone for Germany.
7. Revisit config/countries.json every few months. Salaries change.