CV Overleaf Agent

An agent template for turning a Chinese resume PDF or text resume into an English CV and Overleaf-ready LaTeX.

The workflow is designed for students and early-career candidates who want a clean, focused CV without learning LaTeX first. It preserves the spirit of concise Chinese resume templates such as "职途": clear sections, strong bullet points, compact layout, and no decorative clutter.

What It Does

• Reads a Chinese resume PDF or pasted resume text.
• Extracts education, experience, projects, skills, awards, and contact details.
• Rewrites the content into polished English CV language.
• Generates LaTeX code that can be pasted into Overleaf.
• Guides the user through Overleaf project creation, code replacement, recompilation, and final checks.

Quick Start

Python Agent Loop

Install the project:

pip install -e .

Set your OpenAI API key:

export OPENAI_API_KEY="your-api-key"

Run the agent:

python3 -m overleaf_agent --input resume.pdf --target "data analyst internship" --out outputs

The agent writes:

• outputs/facts.json
• outputs/cv_draft.md
• outputs/resume.tex
• outputs/overleaf_steps.md

You can also use text input:

python3 -m overleaf_agent --input resume.md --target "master's application" --out outputs

Custom GPT Version

1. Open ChatGPT or your preferred agent builder.
2. Copy the instructions from agent/custom-gpt-instructions.md.
3. Upload a Chinese resume PDF or paste resume text.
4. Ask the agent to first produce the English CV content.
5. Ask the agent to generate Overleaf-ready LaTeX.
6. Open Overleaf, create a new blank project, delete the starter code, paste the LaTeX, then click Recompile.

Recommended User Flow

Step 1: Upload Resume

Use this prompt:

I uploaded my Chinese resume PDF. Please extract all useful resume information and organize it into a structured English CV outline. Do not generate LaTeX yet.

Step 2: Generate English CV

Use this prompt:

Please rewrite the extracted content into a polished one-page English CV for international job or graduate-school applications. Keep it concise, achievement-oriented, and faithful to the original facts.

Step 3: Generate LaTeX

Use this prompt:

Please generate complete Overleaf-ready LaTeX code for this English CV. Use a clean one-page layout, compact spacing, and ATS-friendly section names.

Step 4: Use Overleaf

1. Go to Overleaf.
2. Select New Project.
3. Choose Blank Project.
4. Name the project, such as English CV.
5. Delete the default blue starter code.
6. Paste the LaTeX code generated by the agent.
7. Click Recompile.
8. Review the PDF preview and manually adjust details if needed.

Repository Structure

overleaf_agent/
  cli.py                       Command-line entry point
  loop.py                      Agent loop: extract -> draft -> latex -> review -> save
  llm.py                       OpenAI Responses API wrapper
  pdf.py                       PDF/text input reader
  prompts.py                   Prompt templates
  latex.py                     LaTeX cleanup and validation helpers
  files.py                     Output file helpers
tests/
  test_latex.py
  test_loop.py
  test_input_reader.py
agent/
  custom-gpt-instructions.md  Custom GPT builder instructions
  system-prompt.md            Portable system prompt for other agents
  user-prompts.md             Copyable prompt sequence for users
templates/
  resume.tex                  Clean Overleaf LaTeX template
  resume-cn-to-en-notes.md    Translation and editing rules
examples/
  input-brief.md              Example source resume summary
  output-latex.tex            Example generated LaTeX
docs/
  overleaf-workflow.md        Detailed Overleaf guide
  quality-checklist.md        Final review checklist

Output Standards

The agent should produce CVs that are:

• Fact-faithful: never invent school names, dates, employers, awards, metrics, or technologies.
• Concise: prioritize one page unless the user explicitly asks for two.
• English-native: use natural resume verbs, not literal machine translation.
• ATS-friendly: avoid icons, images, progress bars, and unusual section labels.
• Overleaf-ready: return complete LaTeX, not fragments.

License

MIT