This repository contains a GenAI application designed to optimize and tailor CVs.
All statements are supported by links and are not merely personal opinions.
As of 2025, the highest rejection rate (98%) occurs not during recruiter, behavioral, or technical interviews, but at the CV screening stage.
On average, a candidate needs to submit 51 resumes to secure a job. Key reasons include:
- ATS (Applicant Tracking Systems) – Each company has unique configurations, expectations, and requirements. Some demand a 99% match, expecting candidates to have worked on nearly identical projects, while others value passion and minimal coding experience. In reality, 98% of large companies use ATS tools, and 75% of resumes are rejected before reaching a human reviewer.
- Recruiters – Human reviewers are often even less consistent than ATSs. Frequently, after about 10 seconds, a recruiter may decide “other candidates’ experience aligns more closely.”
Companies like Amazon and Google recommend tailoring your resume to each specific job description.
I agree with this approach, although it can feel unfair that sophisticated systems can reject candidates in seconds while job seekers must spend significant time customizing their resumes. To address this, I developed a tool that automates resume tailoring.
Good news - it’s almost free to use.
As of September 2025, I’ve primarily been using GPT-5-mini or GPT-5, which works exceptionally well for this task.
Processing ~240,000 tokens for 100 requests costs around $0.81, making the cost per resume less than $0.01.
The tool allows full customization - you can make your CV polite and professional, follow Big Tech guidelines, or even adopt a more creative “pirate” style that makes it look like you’ve been in a role for years. The choice is yours, though we recommend staying mostly honest.
The tool is distributed as a Python package, offering two usage options:
The simplest way to use the tool is via Docker:
docker run --platform linux/amd64 -p 8501:8501 ghcr.io/vitomakarevich/cv-tuner:latestThen open the local web page to use the app.
The application runs fully locally and does not store any of your data.
However, note that the AI provider may process and store your input.
- UI instruction manual
- You can download generated data directly from the UI or access the full debug version at streamlit_output.
You can also run the tool programmatically using a test script.
Provide your real experience, job description, and configure rules in the template folder.
The tool will process this data and generate the output in output.