Konicamaheshwarii/job-autoapply：自动监控招聘群并投递的 WhatsApp 求职机器人

Job Auto-Apply

A WhatsApp bot that watches job-opening groups, scores every post against your CV, builds a tailored ATS-friendly CV for the good matches and applies by email or WhatsApp. It runs unattended on a home PC.

Highlights

AI only where it pays off. Text posts are parsed with rules, using no tokens. A vision LLM is called only for image posters and for tailoring CVs that will actually be sent.
Multi-provider LLM fallback. A configurable provider:model chain (Groq → Gemini) with retries and daily-quota detection, so free-tier limits never stop the bot.
ATS-style scoring. 70% skill-keyword coverage plus 30% role fit, with skill synonyms (e.g. nodejs = Node.js, restful = REST API), checks on stack mismatch and years of experience, and de-duplication against past applications.
Honest tailoring. Bullets are reordered and filtered, never invented. CV content stays word-for-word from the master CV, and a sanitizer drops any skill, number or seniority claim the model adds.
Self-healing. Watchdogs re-scan groups, restart after PC sleep or a lost connection, and recover when WhatsApp Web stalls. A daily cap and random delays keep activity human-paced.
Dry-run first. DRY_RUN=true runs the whole pipeline, including the PDF, without sending anything.

How it works

flowchart LR
    A[WhatsApp group post] --> B{Text or image?}
    B -- text --> C[Rule parser<br/>no AI]
    B -- image --> D[Vision LLM<br/>extract jobs]
    C --> E[Score vs CV<br/>keywords + role fit]
    D --> E
    E -- below MIN_SCORE / duplicate / wrong stack --> X[Skip + log]
    E -- match --> F[Tailor CV<br/>LLM or rules fallback]
    F --> G[Render ATS PDF<br/>Puppeteer]
    G --> H{Contact}
    H -- email --> I[Gmail / Nodemailer]
    H -- phone --> J[WhatsApp DM]
    I --> K[Notify self + CSV log]
    J --> K

Module	Responsibility
`src/index.js`	WhatsApp client, group backfill, watchdogs, graceful shutdown
`src/pipeline.js`	Orchestrates one post end to end
`src/parse.js` / `src/extract.js`	Rule-based parsing / LLM extraction for posters
`src/ats.js` / `src/match.js`	Keyword coverage, role fit, experience rules
`src/ai.js`	Provider-agnostic JSON LLM client with fallback chain
`src/tailor.js` / `src/tailor-rules.js`	LLM tailoring with sanitizing, plus a rules-only fallback
`src/pdf.js`	HTML → single-column ATS PDF
`src/mailer.js` / `src/store.js`	Gmail sending, application history and daily caps

Setup

Requires Node.js 20+ on Windows, macOS or Linux.

npm install
cp .env.example .env                               # then fill it in
cp cv/master-cv.example.json cv/master-cv.json     # then put your real CV in it

Key	What
`GROQ_API_KEY`	Free key from https://console.groq.com/keys (primary AI)
`GEMINI_API_KEY`	Free key from https://aistudio.google.com/apikey (backup AI)
`WHATSAPP_GROUPS`	Exact group name(s), comma-separated
`GMAIL_USER` / `GMAIL_APP_PASSWORD`	Gmail address + App Password
`DRY_RUN`	`true` = do everything except send. Start with this.
`MIN_SCORE`, `MAX_APPLICATIONS_PER_DAY`	Match threshold (default 70) and daily cap (default 12)

Run

npm run try -- sample-job.txt   # test one post, no WhatsApp needed
npm start                       # scan the QR once (WhatsApp > Linked devices)

On Windows, scripts/run-bot.cmd starts the bot and restarts it if it exits. To run it at login, put a shortcut to it in shell:startup.

Notes

patches/ holds fixes for whatsapp-web.js 1.34.7, which broke after the July 2026 WhatsApp Web update. They are applied automatically after npm install.
WhatsApp does not allow bots on personal accounts. Use at your own risk.
cv/master-cv.json, .env, data/ and out/ are git-ignored because they contain personal data.

License

MIT