ApplyKit

Paste a job URL. Get a tailored cover letter and packaged application documents in seconds.

---

Why this is different from AI-generated cover letters

Most AI cover letter tools generate the entire letter from scratch — resulting in generic, interchangeable output that reads like it was written by an AI.

This tool works differently. You write the cover letter once, optimised to your voice, experience, and strengths. The AI only fills in the small parts that need to change per application — the company name, what draws you to the role, why this specific organisation. Everything else stays exactly as you wrote it.

The result is a letter that sounds like you, not like ChatGPT, because it mostly is you.

---

What it does

1. Scrapes the job posting (title, company, country, responsibilities, qualifications)
2. Auto-detects the job country and selects the right resume/cover letter template
3. Classifies the role by category (Finance, Marketing, etc.) and picks the matching template
4. Fills only the _ blanks in your cover letter with company- and role-specific language
5. Converts to PDF and opens the output folder

---

Setup

Tip: The PROJECT_MEMORY.md file in this repo is a full technical briefing of how the project works. If you get stuck at any point during setup, paste it into Claude, ChatGPT, or any AI assistant and describe your issue — it has everything the AI needs to help you configure, debug, or extend the tool.

1. Install dependencies

pip install -r requirements.txt
playwright install msedge

Windows only: docx2pdf requires Microsoft Word to be installed for PDF conversion.

2. Set your API key

cp .env.example .env

Edit .env and paste your LLM API key.

3. Configure paths

cp config.example.py config.py

Edit config.py and set all paths:

• OUTPUT_BASE — folder where generated applications are saved
• TEMPLATE_BASE — folder containing your resume/cover letter template subfolders
• SUPPLEMENTARY_FILES — any extra PDFs to upload alongside your resume (transcripts, references, etc.)
• BUNDLE_APPENDIX — PDFs to append after the cover letter into a combined bundle PDF (leave empty [] if not needed)
• PROFILES — if applying to multiple countries, define one profile per country with its own OUTPUT_BASE and TEMPLATE_BASE

4. Add your candidate profile

cp applicant.example.json applicant.json

Edit applicant.json with your personal info, work history, education, and skills.

5. Set up your keyword map

cp keywords.example.json keywords.json

Edit keywords.json so each key matches a subfolder name inside your TEMPLATE_BASE. Keywords are matched against the job title (3x weight) and description (1x weight) to pick the right template.

Add a "_broad_categories" key listing any categories that are general fallbacks (e.g. Finance, Accounting). If a broad category wins with no title evidence but a specialist category has description signal, the specialist wins instead. Any category not listed in "_broad_categories" is treated as a specialist.

"_broad_categories": ["Finance", "Accounting", "Investment"]

6. Set up your location map (multi-country only)

cp locations.example.json locations.json

Edit locations.json so each key matches a profile name in PROFILES. Values are city/country strings matched against the scraped job location.

---

Template folder structure

Your TEMPLATE_BASE folder should contain one subfolder per job category:

Templates/
├── Finance/
│   ├── Resume.docx
│   ├── Resume.pdf
│   ├── Resume.txt          ← plain text version of your resume
│   └── Cover Letter.docx
├── Marketing/
│   └── ...
└── ...

Subfolder names must match the keys in keywords.json.

Writing your cover letter template

Write your cover letter in full — your background, experience, skills, and voice. Use _ as a placeholder only where company- or role-specific content belongs. Only those sentences are sent to the AI. Everything else is untouched.

Template:

I am writing to express my keen interest in the Financial Analyst role at _.

What draws me to _ is its reputation for disciplined execution and a culture where data-driven analysis shapes business decisions.

Generated:

I am writing to express my keen interest in the Financial Analyst role at RHB.

What draws me to RHB is its position as one of Southeast Asia's leading financial services groups, where rigorous analysis directly informs strategic decisions.

The AI fills the blank and nothing else. Your sentences stay your sentences.

---

Usage

Run from a terminal (Command Prompt, PowerShell, or bash — not by double-clicking the file):

python apply.py

Paste the job URL when prompted. The tool auto-detects the country, classifies the role, then asks you to confirm the title, company, and category before generating your application.

The browser runs headless (no window appears) during scraping.

---

Single vs multi-country setup

Single country — set OUTPUT_BASE and TEMPLATE_BASE directly in config.py. No country prompt appears.

Multiple countries — define PROFILES in config.py and populate locations.json. Country is auto-detected from the job posting; the prompt only appears if detection fails.

# config.py
PROFILES = {
    "United States": {
        "OUTPUT_BASE":   r"C:\...\Applications\US",
        "TEMPLATE_BASE": r"C:\...\Templates\US",
    },
    "United Kingdom": {
        "OUTPUT_BASE":   r"C:\...\Applications\UK",
        "TEMPLATE_BASE": r"C:\...\Templates\UK",
    },
}

// locations.json
{
  "United States": ["united states", "new york", "san francisco", "chicago"],
  "United Kingdom": ["united kingdom", "london", "manchester", "edinburgh"]
}

---

Personalisation summary

File	What to configure
config.py	All folder paths, country profiles, bundle files
applicant.json	Personal info, work history, education, skills
keywords.json	Job categories and their matching keywords
locations.json	Country/city strings mapped to each profile
Cover Letter.docx	Your template — write it fully, use _ only for company/role-specific sentences

---

Requirements

• Python 3.9+
• Windows (for docx2pdf via Microsoft Word)
• Microsoft Edge (used by Playwright for scraping)
• LLM API key (default: Anthropic Claude — get one here)