How to build a Resume matcher to increase the probability of passing an ATS system with huggingface pipelines

I want to build a pipeline that uses hugging face models and give it two texts one for the job description i want to apply to and the other is for my resume.
is the cosine similarity that metric that i should use? then i should try to get the words or phrases from the job_description and find their way to replace them from the my_resume like:
replace “python developer” into “Python/Django developer” or “Software engineer(SE) with python” ?
how to do this ! i tried gpt and other llama but didn’t find a good result even the cosine similiarty i didn’t find it a real metric to try to improve on.
Here is a prompt that i used a cohere AI model but didn’t find the result as i need it was best than normal gpt 3.5 from openai