Job Application Agent — Project Index

TL;DR

Fully automated job application pipeline running on n8n + local Ollama (Qwen 2.5 7B) + Playwright. While I sleep, robots search 5 job sites every 4–6 hours, score each job against my profile using AI, generate a tailored CV + cover letter for matches scoring 70+, and submit applications automatically.

Cost: ₹0/month. Daily effort: zero (except clicking submit on Tier 1 manual reviews).

What it does (the 5 robots)

1. Scrapers — "The Spies"

Every 4-6 hours, 5 scrapers wake up:

LinkedIn — Customer Success / Project Coordinator / Operations / BDM (India + EU)
Naukri — Same roles for Delhi NCR
Indeed — India + EU
Welcome to the Jungle — European tech roles
Tier 1 career pages — Capgemini, Accenture, JLL, CBRE, EY

Dedupes against memory so it doesn't re-process old jobs.

2. Fit Scorer — "The Judge"

Qwen 2.5 7B Instruct on local Ollama (localhost:11434) scores each job 0-100:

| Criterion | Points | |---|---| | Role title match | 25 | | Skills match | 25 | | Experience match (3 yrs) | 15 | | Salary 10+ LPA | 15 | | AI/tech relevance | 10 | | Growth / EU path | 10 |

HARD FLOOR: 10 LPA. Below = instant REJECT. Threshold: 70+ to apply.

3. CV Builder — "The Tailor"

Uses CV variant A (operations) or B (AI/automation) based on role type
Rewrites Professional Summary to mirror JD language
Reorders bullet points to surface relevant experience first
Never invents experience — only rewrites what's already in the base CV
Generates 250-word cover letter in my voice (direct, not desperate)

4. Executor Agents — "The Submitters"

India Agent → Naukri, Indeed India, LinkedIn India (full auto)
Abroad Agent → LinkedIn EU, WTTJ, Indeed EU (full auto with cover letter)
Tier 1 Agent → Capgemini/Accenture/JLL/CBRE/EY: fills form + screenshot + STOPS for my manual submit

5. App Logger — "The Notebook"

Writes every action to Google Sheet: timestamp, company, title, platform, score, CV variant, status, URL, notes.

Architecture

Scrapers (every 4-6h)
    ↓
Fit Scorer (Qwen 2.5, score 0-100)
    ↓ (70+ only)
CV Builder (tailor CV + write cover letter)
    ↓
Router (India / Abroad / Tier1)
    ↓
Executor Agents (Playwright auto-apply)
    ↓
App Logger (Google Sheets)
    ↓
Local HTML Dashboard (auto-refresh 60s)

My Profile Used by the AI

Name: Debu Rajput (Dev)
Education: MBA from Ambedkar University Delhi (2021–2023)
Experience: 3+ years operations & client management
Background: NON-TECHNICAL (operations, stakeholder management, project coordination)
AI projects (hobbyist): JARVIS (Ollama+Python+LangChain), n8n workflows, LangGraph experiments
Goal: Europe relocation (Netherlands, Germany, UK preferred)
Min salary: 10 LPA hard floor

Target roles (priority order)

AI Operations / AI Customer Success / AI Project Coordinator (NON-CODING AI)
AI Implementation Specialist / AI Solutions Coordinator
Customer Success Manager (AI/SaaS products)
Project Coordinator at AI/tech companies
Operations Manager at AI/tech firms
BDM / Business Analyst at consulting/tech

Reject roles

Pure software engineer / data scientist / ML engineer (I'm non-technical)
Real estate sales, cold calling, telesales, data entry
Anything below 10 LPA

Key metrics in every CV (never remove)

35% faster query resolution
20% CSAT improvement
97% event satisfaction (90+ attendees)
Zero possession handover rejections
20% HR escalation reduction (Conscient)
300+ residential units managed

Tech Stack

| Component | Tool | Cost | |---|---|---| | Workflow engine | n8n (local, port 5678) | Free | | AI model | Qwen 2.5 7B Instruct via Ollama (port 11434) | Free | | Browser automation | Playwright + Chromium | Free | | Application log | Google Sheets | Free | | Dashboard | Local HTML reading Google Sheet CSV | Free |

Originally built on Anthropic Claude API → migrated to local Ollama for zero cost + privacy.

File Reference

job-agent/
├── HOW_IT_WORKS.md                  # Plain English explanation of all 5 robots
├── SETUP_GUIDE.md                   # Step-by-step walkthrough
├── README.md                        # Technical setup
├── .env                             # Local config (OLLAMA_MODEL, MIN_SALARY_LPA, etc.)
├── cv/
│   ├── cv_variant_a.md              # Operations-heavy CV (default)
│   ├── cv_variant_b.md              # AI/Automation-heavy CV
│   └── master_profile.md            # Full profile for AI context (don't modify)
├── prompts/
│   ├── fit_scorer.md                # AI Judge criteria + hard 10 LPA filter
│   ├── cv_tailor.md                 # CV tailoring rules
│   └── cover_letter.md              # My voice for cover letters
├── core/
│   ├── fit_scorer.json              # n8n: webhook /score-job
│   ├── cv_builder.json              # n8n: webhook /build-cv
│   └── app_logger.json              # n8n: webhook /log-application
├── agents/
│   ├── india_agent.json             # n8n: webhook /apply-india
│   ├── abroad_agent.json            # n8n: webhook /apply-abroad
│   └── tier1_agent.json             # n8n: webhook /apply-tier1 (manual review)
├── scrapers/
│   ├── linkedin_rss.json            # every 4h
│   ├── naukri_scraper.json          # every 4h
│   ├── indeed_scraper.json          # every 4h
│   ├── wttj_scraper.json            # every 6h
│   └── tier1_scraper.json           # every 6h
├── playwright/
│   ├── linkedin_apply.js            # LinkedIn Easy Apply
│   ├── naukri_apply.js              # Naukri apply form
│   └── tier1_form_fill.js           # Tier 1 ATS (no auto-submit)
├── dashboard/
│   └── index.html                   # Live dashboard reading Google Sheet
├── setup/
│   ├── install.sh                   # Full install
│   ├── test_scorer.js               # Verify AI scorer end-to-end
│   ├── FIX_AUTOSTART_ONCE.bat       # One-time admin fix for SoulInMotion-n8n RunLevel
│   └── KILL_AND_RESTART_N8N.bat     # Manual restart helper
└── output/                          # Generated CVs, cover letters, screenshots

Critical Operational Notes

Soul in Motion + Job Agent share the same n8n

Soul in Motion's SoulInMotion-n8n scheduled task auto-starts n8n at boot
Both projects' workflows live in the same n8n SQLite DB
Restarting n8n briefly pauses Soul in Motion (~10s) but no data is lost
Original task ran with elevated privileges (locked port 5678 from non-admin sessions)
Fixed by setup/FIX_AUTOSTART_ONCE.bat (downgrades RunLevel from Highest → Limited)
After running once, future n8n bounces work without admin

What I have to do daily

Nothing. Except:

Click "Submit" on Tier 1 forms (Capgemini/Accenture/JLL/CBRE/EY) after reviewing screenshot
Open dashboard to see what was applied to overnight

What I edit when I want to tune behavior

cv/cv_variant_a.md / cv_variant_b.md — my CVs
prompts/fit_scorer.md — what the Judge looks for
prompts/cover_letter.md — my voice
.env — MIN_SALARY_LPA and FIT_THRESHOLD

Setup Status (as of 2026-05-05)

| Step | Status | |---|---| | Folder structure created | ✅ Done | | All 11 n8n workflows built | ✅ Done | | All 3 Playwright scripts built | ✅ Done | | 2 CV variants written | ✅ Done | | Ollama + Qwen 2.5 7B installed | ✅ Done | | Workflows imported to n8n via CLI | ✅ Done | | Workflows activated in n8n DB | ✅ Done | | Pushed to GitHub | ✅ Done | | Dashboard built | ✅ Done | | n8n auto-start fixed (run FIX_AUTOSTART_ONCE.bat) | ⏳ Pending | | Google Sheet created + headers | ⏳ Pending | | Dashboard linked to Sheet ID | ⏳ Pending | | Google Sheets service account credentials | ⏳ Future | | 15-day no-reply follow-up watcher | ⏳ Future |

Future Iterations

Google Sheets service account credentials — so n8n can WRITE (not just dashboard READ)
15-day no-reply follow-up watcher — auto-nudge after 15 days no reply
WhatsApp/Email notification — ping when Tier 1 form is ready for manual submit
Glassdoor scraper — additional pipeline source

Related Knowledge

[[ai-agents-learning/_index]] — broader AI agent architecture notes
[[europe-relocation/_index]] — Europe relocation strategy and target companies
[[real-estate-automation/_index]] — current job context (will be replaced when relocated)
[[../personal-brand/_index]] — LinkedIn presence (Soul in Motion)

Build Log

2026-05-05 — Initial build (all 26 files, ~2,500 lines) 2026-05-05 — Migrated Anthropic API → local Ollama (Qwen 2.5 7B) 2026-05-05 — Tightened filters: 10 LPA min, 70+ threshold, AI non-coding focus 2026-05-05 — Added live HTML dashboard 2026-05-05 — Added FIX_AUTOSTART_ONCE.bat (Soul in Motion task RunLevel fix) 2026-05-05 — Added HOW_IT_WORKS.md + SETUP_GUIDE.md

JOB-APPLICATION-AGENT