linkedin-auto-apply

Automated LinkedIn job applicant powered by a headless Chromium browser and a local LLM.

Replicates the full auto-apply/ bot natively using kdeps browser resources: logs in, searches jobs with filters, scores each listing against your profile, and runs the complete Easy Apply form flow (fill questions, upload resume, submit) for every matched job.

---

Pipeline

Step	Resource	What it does
1	load-profile	Parse YAML front matter from the file input source
2	set-profile	Populate kdeps memory from parsed profile JSON
3	validate	Check all required inputs
4	login	Chromium logs in to LinkedIn, session persisted
5	search-jobs	Navigate job search, extract listings via JS
6	analyze-jobs	LLM scores each job, filters by threshold, writes cover letter
7	apply-job	Per matched job: click Easy Apply, fill form, upload resume, submit
8	response	Return applied/skipped/failed summary

---

Prerequisites

• kdeps installed (kdeps --version)
• Ollama running locally with llama3.2

  ollama pull llama3.2
  

• A LinkedIn account with Easy Apply access
• Your resume as a local PDF file

---

Profile file

Copy sample-profile.md to joel-profile.md (or any name) and fill in your details. The file uses YAML front matter between --- delimiters:

---
job_title: "Senior Software Engineer"
location: "Amsterdam, Netherlands"
candidate_name: "Jane Doe"
linkedin_email: "jane@example.com"
linkedin_password: "your-password"
resume_path: "/Users/jane/resume.pdf"
# ... see sample-profile.md for all fields
---

---

Run

Pass the profile file via --file, stdin, or KDEPS_FILE_PATH:

# --file flag (recommended)
kdeps run examples/linkedin-auto-apply --file joel-profile.md

# stdin pipe
cat joel-profile.md | kdeps run examples/linkedin-auto-apply

# environment variable
KDEPS_FILE_PATH=joel-profile.md kdeps run examples/linkedin-auto-apply

The workflow reads the markdown file, parses the YAML front matter, and runs the full browser automation pipeline without any API server or HTTP POST required.

---

How it compares to the Selenium bot

Feature	Selenium bot (auto-apply/)	This workflow
Login	Selenium + credentials	Playwright/Chromium + credentials
Job search	driver.get(search URL)	Browser resource navigate
Filters	Click UI filter elements	URL params (f_TPR, f_WT, f_LF)
Job scoring	Not built-in	LLM scores 0-100 against profile
Cover letter	Static text from config	LLM generates per-job
Easy Apply	Selenium click-through	Browser evaluate JS
Form filling	Label-match + config values	Label-match + config values (same logic)
Resume upload	input.send_keys(path)	Browser upload action
Submit	wait_span_click("Submit")	JS click submit button
Already-applied check	Job state footer text	JS footer text check
Session	Selenium WebDriver	Playwright sessionId
Profile input	Python config files	File input source (--file / stdin)

---

Tuning

Use a larger model for better job scoring in workflow.yaml:

model: deepseek-r1

Lower the score threshold in your profile file:

min_match_score: "40"

Remote jobs only:

remote_only: "true"