I Built a Complete AI Resume with 90+ ATS Score

A-T-S!! A hurdle that most applicants just can’t cross. Spending hours on Overleaf and resume making websites to create their perfect resumes: just to find out that it has an ATS score of 40! For some it’s a dead end. Regardless of what they try, the score doesn’t seem to get anywhere near the required limit (80+). 

What to optimize? Or more specifically how to optimize? Lack of information about optimizing resumes further adds to the problem.

This guide is here to solve that problem. Contained within are guidelines, tips and techniques that you can use with an AI, to boost the ATS score of your resume. 

Why is ATS important?

An applicant tracking system (ATS) is software that companies use to manage hiring, from posting jobs to screening and organizing applications. It helps recruiters handle large volumes of resumes quickly, automate routine tasks, and track candidates through each stage of the process. 

Because many employers rely on ATS tools to filter and rank applicants, job seekers often optimize their resumes with clear formatting and relevant keywords so their applications are accurately parsed and not overlooked before a human ever reviews them.

How to improve ATS score?

Things that work for ATS: 

• Shouldn’t have spelling mistakes
• Uses consistent styling
• Correct text emphasis—Doesn’t effect ATS but important for human recruiters
• Right sections at the top which is:
  • Experience/Work history for experienced individual
  • Education/Projects for freshers
• Has a word count of about 400 words (experience/project section)
• Use a format that is parsable via ATS therefore:
  • No double columns
  • No fancy images
  • No eccentric fonts
• Has reachout options like phone number, email, LinkedIn etc. (preferably clickable). 

These pointers will be used when we’re making the prompt for optimizing our resume. If you know any more enhancement tips, make sure to add them to this list (also share them in the comments for others). 

Tools Used

For demonstration purposes I’ll be using the ATS checker from Weekday. This is one of my favorite ATS checkers, as it’s one of the only ATS checkers that checks the score for a specific position, instead of giving a global score of the resume—which almost never works. It not only outlines the score, but also gives feedback on the improvements that could be made.

This we’d use to our benefit when we would make AI improve our resume based on its feedback.

For the choice of Model, I’d be using ChatGPT. Based on my experience, ChatGPT gives the best response when it comes to following a set of guidelines for curating a resume. Plus it’s really good at making changes based on the provided feedback. 

Getting the Information

Firstly, I’d be requiring qualifications, skills, education etc. that I would put in a resume. This information can either be entered manually, or could be sourced from the profile page of our LinkedIn account. I’d be using the information present in a data scientist’s profile for this:

████████████████  V A S U   D E O  ████████████████

Ghaziabad, Uttar Pradesh
+91-212773314 | [email protected]
https://www.linkedin.com/in/handshake123
github.com/vasdevvv


══════════════════  EXPERIENCE  ══════════════════

ASSOCIATE DATA SCIENTIST | Pluto.ai - June 2025 – Present | Bengaluru, Karnataka

Juggernaut: A Content Management System
Tech Stack: Python, LangChain, LangGraph, LangSmith, RAG, Qdrant,
            OpenAI (GPT-4-mini, text-embedding-3-small),
            Anthropic (Claude 4.5 Haiku), Django, Docker, PostgreSQL

• Built an agentic assistant using LangGraph to orchestrate a multi-LLM backend with a Qdrant-based RAG pipeline, ensuring safe and context-aware routing with structured guardrails.
• Implemented a scalable quiz generation pipeline with context-engineered prompting and used LangSmith for monitoring, tracing, and debugging.

Manual: AI-Powered Evaluation System
Tech Stack: Python, LangChain, LangGraph, LangSmith, RAG, Qdrant,
            OpenAI (GPT-4.1-mini), LLM as a Judge

• Developed automated grading systems to evaluate RAG pipelines and agentic workflows for domain-specific use cases.
• Defined evaluation metrics for response quality, retrieval performance, and agent decision accuracy.

Recommendation System
Tech Stack: Python, OpenAI text-embedding-3-small,
            scikit-learn (TF-IDF), ChromaDB

• Increased user engagement by 24% by deploying a hybrid recommendation engine combining semantic search and keyword matching.
• Engineered an end-to-end data pipeline from external data collection to embedding storage and validation.


DATA SCIENCE INTERN | Pluto.ai - February 2025 – May 2025 | Bengaluru, Karnataka

AI-Powered Call Analysis Platform
Tech Stack: Python, AssemblyAI, Whisper AI, Mistral, Pydub

• Developed a call analysis platform delivering data-driven performance insights.
• Engineered a pipeline for audio processing, transcription, and conversational analysis.


══════════════════  PROJECTS  ══════════════════

• Conversational AI with LangGraph: Engineered a conversational AI agent using Django and LangGraph with real-time chat and multi-step task execution.
• Diabetic Retinopathy Detection: Implemented transfer learning models with an interactive interface for medical image analysis.
• Alzheimer’s Detection: Built an ensemble ML model achieving high classification accuracy on MRI-derived data.

══════════════════  TECHNICAL SKILLS  ══════════════════

Programming Languages / Frameworks: Python, Java, Django, Flask, FastAPI
Generative AI: Prompt Engineering, RAG, LangChain, LangGraph, CrewAI,
Finetuning (PEFT), Stable Diffusion
Data Science: SQL, Pandas, scikit-learn, Keras, PyTorch, XGBoost,
LightGBM, NLP, Computer Vision, GANs

═════════════════════  EDUCATION  ════════════════════

Dr. APJ Abdul Kalam Technical University
B.Tech in Computer Science with Data Science
2021 – 2025

Every part of the above information can be altered to tailor the prompt to yourself. Fill it with your own details. 

Creating the Prompt

Use the above information and create a resume in PDF format that scores over 90 in ATS checkers. Here are some guidelines that you should keep in mind while creating the resume:

• Use Arial font
• Uses consistent styling
• Correct text emphasis—Bold, italics when needed
• Right sections at the top which is:
  ◦ Experience/Work history for experienced individual
  ◦ Education/Projects for freshers
• Has a word count of about 400 words (experience/project section)
• Use a format that is parsable via ATS therefore:
  ◦ No double columns
  ◦ No fancy images
  ◦ No eccentric fonts
• Has reachout options like phone number, email, LinkedIn etc. (preferably clickable).
• And use other guidelines that you can find over the internet, that would help in improving the ATS score for the resume.

Response:

Based on these, the model gave a resume in response, with an ATS score of 78. 

This is a good start. Now we can use the feedback provided at the end of the ATS evaluation and provide that as input to ChatGPT for it to improve these things in the resume and send a refined version. 

Updated Prompt:

Suggestions to improve your resume score

• Add full URLs for LinkedIn and GitHub profiles to improve ATS parsing.
• Specify exact startMonth and endMonth for all education and job entries; currently missing for Education and Data Science Intern role.
• Provide a detailed description for the B.Tech degree including relevant coursework or projects to enhance keyword matching.
• Include location details consistently for all experiences; Data Science Intern location is present but could be standardized.
• Expand on certifications by adding issuing organizations, dates, and credential IDs if available.
• Add contact phone number in international format without dashes to improve ATS recognition.
• Use consistent date formats throughout the resume, e.g., replace “June 2025 - Present” with “06/2025 - Present”.
• Incorporate more keywords from the job description you are targeting, such as specific tools or methodologies relevant to the role.
• Include a professional summary or objective section at the beginning with targeted keywords.
• Add measurable achievements or metrics for each project to quantify impact.

Response:

Wow! We were able to get the ATS score of the resume to over 80 in just a single feedback loop. But there is still scope for improvement. 

The 5% Rule

Now, I can buff the score up to 90 or even above, doing the same technique over and over again. But that will make the resume robotic. It will breeze past the ATS checkers but will get immediately rejected by the human recruiters. So, for the last 5 percent, we’d be making some changes in the resume ourselves. This would assure that it has a humane touch. 

All this will be done by providing an enhanced prompt to ChatGPT. But instead of using the feedback from ATS checker, we’d be putting in the things that we feel is missing in the resume.

Here are some things that I felt was missing or could be improved upon on the previous resume:

• Improve the presentation of the information
• Use bold to highlight sought after technologies and frameworks 
• Instead of having a separate section for contact, include that in the header and center the header
• The experience section should follow the summary section
• The headers should be in all caps
• Include links for Github, LinkedIn, Email etc. in raw format as that would improve parsing
• Remove the comma between the month and year in dates
• Use duotone for headers

After providing the previous points in the next prompt, here’s the score for the resume:

95% ATS Score with almost no human intervention. If there is still something that isn’t to your liking, you can go ahead and edit some parts of the resume yourself on an online pdf editor like smallpdf.

Note: Some parts of the resume were truncated to keep the text intact when viewing. Section like education and certification even though don’t show up in the images, do exists at the end of the resume.

Grounding Expectations

One important thing you should know: ATS checkers reward keyword stuffing and fake precision. The guidelines and the prompts provided previously will boost your ATS score, but that resume would lack personification that companies look for. So make sure you give it a final edit and add unique experiences that would help distinguish your resume from typical AI slop. 

Also, you need to recognise when the feedback is useful and when it isn’t. For example, in the last image, the 2^nd feedback was useless. As adding a https: before the URL doesn’t make any difference in parsing. You have to make a call for figuring out these wrong suggestions.

Also a resume with a high ATS score won’t help you with:

• Securing ghost openings
• Getting a robotic resume accepted by a human recruiter

Further Improvements

This technique is primarily based on feedback loops and guidelines for ATS. I mentioned the pointers that had worked for me in improving my ATS score. But this list by no means is exhaustive. 

If you feel there are certain points that can be added in the prompt that will further improve the score, please share in the comment section below. If you know someone who is in dire need of getting hired, make sure to share this with them. 

Frequently Asked Questions