Ai Engineer Ats Optimization Checklist

ATS Optimization Checklist for AI Engineer Resumes

The Bureau of Labor Statistics projects 20% employment growth for computer and information research scientists (SOC 15-1221) through 2034—nearly seven times the 3% average across all occupations—with a median annual wage of $140,910 and top earners exceeding $232,120 ¹². Meanwhile, AI-related job postings climbed from 1.4% to 1.8% of all U.S. job postings between 2023 and 2024 according to Stanford's AI Index Report, with Python appearing as the top specialized skill across those listings ³. That surge means more applications per opening, more aggressive ATS keyword filtering—Jobscan's 2025 survey found 99.7% of recruiters use ATS filters to sort candidates, with 76.4% starting their search by filtering on skills ⁴—and more resumes rejected by software before a hiring manager reads a single line about your transformer architecture expertise.

This checklist covers ATS parsing rules, keyword strategies, formatting requirements, and optimization techniques specific to AI engineers working across machine learning, deep learning, NLP, computer vision, generative AI, and MLOps.

Key Takeaways

• Framework-specific keywords determine ATS ranking. PyTorch appears in 37.7% of AI engineering job postings and TensorFlow in 32.9%—listing "deep learning frameworks" without naming them misses both keyword matches ⁵.
• Quantified model performance separates ranked resumes from rejected ones. Inference latency reductions (340ms to 45ms), accuracy improvements (F1 0.72 to 0.91), dataset sizes (2.3M labeled samples), and GPU utilization percentages (78% cluster efficiency) all pass through ATS as searchable text and immediately communicate your impact level to human reviewers.
• MLOps and deployment skills are now table stakes. Docker appears in 15.4% and Kubernetes in 17.6% of AI job postings—candidates who list only research skills without production deployment experience are filtered out of the majority of industry roles ⁵.
• Cloud certifications function as high-signal ATS keywords. Google Professional Machine Learning Engineer and AWS Machine Learning certifications appeared in 40% more job postings than competing credentials in 2025 ⁶.
• Format compliance prevents silent rejection. Tables, two-column layouts, graphics-based skill bars, and content placed in headers or footers cause ATS parsers to scramble field assignments or drop sections entirely—your CUDA optimization work disappears before anyone reads it ⁴.

Common ATS Keywords for AI Engineers

The keywords below are drawn from O*NET task descriptions for SOC 15-1221, analysis of 3,000+ AI engineering job postings ⁵, and current framework and platform documentation ⁷⁸. Organize them by category on your resume rather than listing them in a flat block.

Hard Skills

Programming Languages: Python (71% of postings), C++ (GPU-optimized code), Java (22% of postings), Rust (inference engines), SQL (17.1% of postings), JavaScript/TypeScript (API layers), Go (microservices), Bash/Shell scripting ⁵

Deep Learning Frameworks: PyTorch, TensorFlow, JAX, Keras, ONNX, TensorRT, Hugging Face Transformers, spaCy, scikit-learn, XGBoost, LightGBM

Generative AI & LLM Tools: LangChain, LlamaIndex, Hugging Face (model hub, tokenizers, datasets), OpenAI API, Anthropic Claude API, Retrieval-Augmented Generation (RAG), vector databases (Pinecone, Weaviate, ChromaDB, Milvus, Qdrant), prompt engineering, fine-tuning (LoRA, QLoRA, PEFT), RLHF ⁸

MLOps & Infrastructure: Docker, Kubernetes, MLflow, Kubeflow, Weights & Biases, Ray, Airflow, DVC (Data Version Control), Seldon Core, BentoML, TorchServe, Triton Inference Server, GitHub Actions, Jenkins, Terraform

Cloud Platforms: AWS (SageMaker, Bedrock, Lambda, EC2, S3), Google Cloud (Vertex AI, TPU, BigQuery), Azure (Azure ML, Azure OpenAI Service, Cognitive Services) ⁵

Data Engineering: Apache Spark, Kafka, Snowflake, Databricks, dbt, Pandas, NumPy, Polars, Delta Lake, Feast (feature store)

GPU & Compute: CUDA, cuDNN, NVIDIA A100/H100, distributed training (DeepSpeed, FSDP, Horovod), mixed-precision training (FP16/BF16), model parallelism, data parallelism

Soft Skills

Cross-functional collaboration (product, engineering, data science), technical documentation, research paper implementation, stakeholder communication, experiment design, code review, mentoring junior engineers, Agile/Scrum methodology, technical writing, conference presentation

Industry Terms & Methodologies

Core ML Concepts: Supervised learning, unsupervised learning, reinforcement learning, transfer learning, few-shot learning, zero-shot learning, self-supervised learning, contrastive learning, attention mechanism, transformer architecture, convolutional neural network (CNN), recurrent neural network (RNN), generative adversarial network (GAN), diffusion model, variational autoencoder (VAE)

NLP Terminology: Named entity recognition (NER), sentiment analysis, text classification, question answering, summarization, machine translation, tokenization, embeddings (word2vec, BERT, sentence-transformers), semantic search, intent classification

Computer Vision Terminology: Object detection (YOLO, Faster R-CNN), image segmentation (U-Net, Mask R-CNN), image classification, pose estimation, optical character recognition (OCR), video understanding, 3D reconstruction

Evaluation & Metrics: Precision, recall, F1 score, AUC-ROC, BLEU score, perplexity, inference latency, throughput (tokens/second), model size (parameter count), FLOPS, A/B testing, statistical significance

Resume Format Requirements

ATS parsers read documents sequentially—left to right, top to bottom—and assign content to fields based on section header recognition ⁴. AI engineer resumes face specific parsing risks because technical content often includes code snippets, architecture diagrams, and mathematical notation that ATS cannot interpret.

File Format

Submit as .docx unless the posting explicitly requests PDF. Word documents parse more reliably across all major ATS platforms (Workday, Greenhouse, Lever, iCIMS, Taleo). If PDF is required, export from Word rather than designing in LaTeX or a layout tool—this preserves the underlying text layer that ATS reads. LaTeX-generated PDFs can render correctly for humans but contain font encoding that some ATS parsers misread.

Layout Structure

• Single column only. Two-column layouts cause ATS to interleave left and right content. A sidebar listing Python libraries alongside work history will merge unpredictably.
• No tables, text boxes, or graphics. Engineers frequently use tables to organize framework proficiency grids or architecture diagrams. ATS reads table cells in unpredictable order or skips them entirely.
• No headers or footers for critical content. Your name, credentials, and contact information belong in the document body—25% of ATS platforms ignore header/footer content during parsing ⁹.
• Standard section headings. Use exactly: "Professional Summary," "Professional Experience," "Technical Skills," "Education," "Certifications," "Projects" (optional). Avoid non-standard headings like "ML Arsenal" or "Research Toolkit."
• No code blocks or mathematical notation. ATS cannot parse inline code formatting, LaTeX equations, or Unicode mathematical symbols. Write "trained a 7-billion-parameter transformer model" instead of embedding model architecture notation.

Font and Spacing

Use 10-12pt in a standard font (Calibri, Arial, Times New Roman, Garamond). Minimum 0.5-inch margins. Avoid condensed or monospace fonts. Use bold for section headers and job titles only; avoid italic for critical keywords since some OCR layers misread italic characters.

Name and Credentials Header

Format your name with credentials on the first line of the document body:

SARAH CHEN, MS
AI Engineer | Machine Learning & NLP
sarah.chen@email.com | (555) 234-5678 | linkedin.com/in/sarahchenml | github.com/sarahchen

This ensures ATS captures your specialization in the title field and your GitHub profile as a searchable text string. Including both LinkedIn and GitHub addresses the two platforms that AI engineering recruiters check most frequently.

Professional Experience Optimization

AI engineering achievements become ATS-competitive when they include model metrics, infrastructure scale, dataset sizes, and business impact. Generic descriptions like "built machine learning models" contain no searchable differentiators.

Bullet Formula

[Action verb] + [ML deliverable] + [framework/tool] + [scale metric] + [outcome/impact]

Before and After Examples

1. Model Training - Before: "Trained deep learning models for text classification" - After: "Trained BERT-based text classification model in PyTorch on 1.8M labeled documents, improving F1 score from 0.76 to 0.93 and reducing manual review workload by 340 analyst-hours per month"

2. LLM Deployment - Before: "Deployed language models to production" - After: "Deployed fine-tuned LLaMA 2 13B model on AWS SageMaker with TensorRT optimization, reducing inference latency from 340ms to 45ms per request while serving 12,000 daily active users at 99.7% uptime"

3. RAG Pipeline - Before: "Built a chatbot using AI" - After: "Architected Retrieval-Augmented Generation pipeline using LangChain, Pinecone vector database, and GPT-4, indexing 450K internal documents and achieving 91% answer accuracy on domain-specific queries measured against expert-labeled test set of 2,000 questions"

4. Computer Vision - Before: "Worked on computer vision projects" - After: "Developed YOLOv8-based defect detection system in PyTorch processing 2,400 manufacturing images per hour on NVIDIA A100, achieving 96.2% [email protected] and reducing false positive rate from 8.3% to 1.1%, saving $2.1M annually in manual inspection costs"

5. MLOps Pipeline - Before: "Set up ML infrastructure" - After: "Built end-to-end MLOps pipeline using Kubeflow, MLflow, and GitHub Actions automating model training, evaluation, and deployment across 14 production models, reducing model update cycle from 3 weeks to 48 hours with automated drift detection via Evidently AI"

6. Data Pipeline - Before: "Processed data for machine learning" - After: "Engineered feature pipeline in Apache Spark processing 2.3TB of clickstream data daily, generating 847 features stored in Feast feature store and reducing training data preparation time from 6 hours to 22 minutes"

7. NLP System - Before: "Built NLP models" - After: "Developed multi-language NER system using spaCy and Hugging Face Transformers supporting 8 languages, extracting 23 entity types from 500K clinical documents with 94.7% entity-level F1 and deploying via FastAPI microservice handling 1,200 requests per minute"

8. GPU Optimization - Before: "Optimized model training speed" - After: "Implemented distributed training using PyTorch FSDP across 32 NVIDIA A100 GPUs, reducing training time for 7B-parameter language model from 14 days to 38 hours while achieving 78% GPU cluster utilization through mixed-precision (BF16) training"

9. Recommendation System - Before: "Built recommendation engine" - After: "Designed two-tower neural recommendation model in TensorFlow Serving processing 45M daily user interactions, improving click-through rate by 23% and incremental revenue by $4.8M annually through real-time personalization with sub-50ms P99 latency"

10. Fine-Tuning & Alignment - Before: "Fine-tuned language models" - After: "Fine-tuned Mistral 7B using QLoRA (4-bit quantization) on 85K domain-specific instruction-response pairs, achieving 12-point improvement on internal benchmark while reducing GPU memory requirements from 80GB to 24GB, enabling deployment on single NVIDIA A10G instance at $0.38/hour"

Skills Section Strategy

The skills section serves a dual purpose: keyword density for ATS matching and quick-scan reference for human reviewers. Structure it for both audiences.

Recommended Format

Group skills under 4-5 sub-headers rather than listing them in a single block. This improves both ATS parsing (clear categorization) and readability.

Deep Learning & ML Frameworks: PyTorch, TensorFlow, JAX, Hugging Face Transformers, scikit-learn, XGBoost, ONNX, TensorRT

LLM & Generative AI: LangChain, LlamaIndex, RAG pipelines, vector databases (Pinecone, Weaviate), fine-tuning (LoRA, QLoRA), prompt engineering, RLHF

MLOps & Infrastructure: Docker, Kubernetes, MLflow, Kubeflow, Weights & Biases, Ray, Airflow, GitHub Actions, Terraform

Cloud Platforms: AWS (SageMaker, Bedrock, Lambda), GCP (Vertex AI, TPU), Azure ML

Programming & Data: Python, C++, SQL, Spark, Kafka, Pandas, NumPy, CUDA, Git

Mirror the Job Posting

Read the specific job posting before submitting. If the posting says "Hugging Face," do not write "HF" alone—ATS performs string matching, not conceptual matching. If the posting says "Retrieval-Augmented Generation," use that exact phrase, not "RAG" alone. If it says "large language models," use that term alongside "LLM." Include both the abbreviated and full forms when space allows: "Retrieval-Augmented Generation (RAG)" ⁴.

Certifications as Keywords

List credentials with both the abbreviation and full name on first occurrence. Google Professional ML Engineer and AWS ML certifications appeared in 40% more job postings than competing credentials in 2025 ⁶:

• AWS Certified Machine Learning Engineer - Associate — Attained 2025
• Google Cloud Professional Machine Learning Engineer — Attained 2024
• NVIDIA Certified Associate: Generative AI LLMs — Attained 2025
• DeepLearning.AI Deep Learning Specialization (Coursera) — Completed 2023
• MS in Computer Science, Machine Learning specialization — Stanford University, 2022

This ensures ATS matches whether the recruiter searches "AWS ML," "Machine Learning Engineer," or the full certification name.

Common ATS Mistakes AI Engineers Make

1. Listing Frameworks Without Version or Context

Writing "PyTorch" in a skills list tells ATS you have the keyword, but tells a hiring manager nothing about your depth. "PyTorch 2.0 — 4+ years production use, distributed training (FSDP), custom dataset pipelines, TorchScript model export" provides ATS keywords while communicating proficiency. With deep learning appearing in 28.1% of AI engineering postings, framework context separates your application from candidates who completed a single tutorial ⁵.

2. Omitting Production Scale Metrics

"Built a machine learning model" contains zero differentiating information. How many parameters? What dataset size? What was the latency? What throughput did it handle? A bullet with "trained 3B-parameter model on 500K samples, serving 8,000 requests/minute at 42ms P95 latency" contains eight additional searchable terms and immediately communicates seniority level. Scale metrics are the AI engineering equivalent of revenue numbers—they signal whether you operate at startup or enterprise scale.

3. Using Research Paper Formatting

Academic CVs use LaTeX, multi-column layouts, and dense bibliographies. ATS cannot parse any of these reliably. If you are transitioning from research to industry, rebuild your resume in a single-column Word document with standard section headers. Move your publications list to a simple bulleted format: "First Author, 'Efficient Attention Mechanisms for Long-Context Generation,' NeurIPS 2024" rather than using BibTeX formatting.

4. Confusing ML Research Skills with ML Engineering Skills

Listing "gradient descent," "backpropagation," and "loss function design" signals academic knowledge but not engineering capability. Recruiters searching AI engineering roles filter for deployment terms: "model serving," "CI/CD for ML," "A/B testing," "monitoring," "feature store," "latency optimization." A resume heavy on theory but missing MLOps terminology will be filtered out of 75% of industry postings that specifically seek production-oriented engineers ⁵.

5. Submitting One Resume for All AI Roles

An NLP engineer's keyword profile and a computer vision engineer's keyword profile overlap less than candidates assume. "Tokenization," "attention mechanism," and "BLEU score" are NLP terms. "mAP," "IoU," and "anchor boxes" are CV terms. "MLOps engineer" searches for "Kubernetes," "model registry," and "drift detection." A resume listing all of these dilutes your relevance score for any single posting. Tailor to the specific sub-domain.

6. Burying GitHub and Publications Below Page One

AI engineering hiring managers check GitHub contribution history and publications as primary qualification signals. If your GitHub URL and top publications appear on page two, ATS ranking algorithms that weight earlier-appearing content may deprioritize them. Place GitHub, Google Scholar, and your top 2-3 publications in your contact header area or immediately after your professional summary.

7. Using Graphics for Technical Architecture

System architecture diagrams, model comparison charts, and training curves are invisible to ATS. The system extracts zero text from embedded images. Replace visual representations with descriptive text: "Designed microservice architecture with 3 model-serving endpoints (recommendation, classification, extraction) behind API gateway, processing 45M daily requests across 12 Kubernetes pods with horizontal auto-scaling."

ATS-Friendly Professional Summary Examples

Your professional summary should contain 3-5 sentences packing your highest-value keywords, years of experience, specialization, and production context. ATS weights content appearing earlier in the document more heavily on some platforms ⁴.

Entry-Level: ML Engineer (0-2 Years)

Machine Learning Engineer with 2 years of experience building and deploying deep learning models in PyTorch and TensorFlow. Developed NLP classification pipeline processing 200K documents using Hugging Face Transformers and deployed to AWS SageMaker with Docker containerization, achieving 91% accuracy on production workload. Proficient in Python, SQL, MLflow experiment tracking, and Git-based ML workflows. MS in Computer Science with published research on efficient transformer fine-tuning (AAAI 2025). AWS Certified Machine Learning Engineer - Associate.

Mid-Career: Senior AI Engineer (3-6 Years)

Senior AI Engineer with 5 years of experience designing and deploying production ML systems across NLP, recommendation, and generative AI applications. Led development of RAG-based enterprise search platform using LangChain, Pinecone, and GPT-4 serving 15,000 daily active users at sub-200ms latency. Built end-to-end MLOps pipelines in Kubernetes with MLflow, Airflow, and automated model retraining handling 14 production models. Experienced in PyTorch distributed training across multi-GPU clusters (NVIDIA A100), reducing training costs by 40% through mixed-precision optimization. Google Cloud Professional Machine Learning Engineer.

Senior: Staff AI Engineer / ML Architect (7+ Years)

Staff AI Engineer with 9 years of experience leading ML platform architecture and AI strategy for enterprise-scale systems processing 200M+ daily predictions. Directed team of 12 ML engineers building foundation model infrastructure on AWS (SageMaker, Bedrock) supporting 6 product teams and reducing model deployment time from 4 weeks to 2 days through standardized MLOps tooling. Architected distributed training platform using PyTorch FSDP and Ray across 128 NVIDIA H100 GPUs, training custom 13B-parameter domain model achieving state-of-the-art performance on 3 internal benchmarks. Published 8 papers at NeurIPS, ICML, and ACL with 1,200+ citations. AWS Certified Machine Learning Engineer, NVIDIA Certified Associate: Generative AI LLMs. MS in Computer Science (Machine Learning), Stanford University.

Frequently Asked Questions

Should I list every ML framework and library I have used?

List frameworks and libraries where you have production experience or substantial project work—not every package you imported once. ATS matches keywords regardless of proficiency, but human reviewers will probe your claimed skills in interviews. For high-priority keywords (PyTorch, TensorFlow, Hugging Face Transformers), add brief context: "PyTorch — 4+ years, distributed training, custom model architectures, TorchScript deployment." For secondary tools (pandas, NumPy, matplotlib), a grouped listing without context is sufficient. Prioritize the tools that appear in the specific job posting you are targeting ⁴⁵.

How do I handle the ML Research vs. ML Engineering distinction on my resume?

Be explicit about which hat you wear. If the posting says "ML Engineer," lead with deployment and production metrics: models served, latency, throughput, uptime, and infrastructure scale. Position research experience as supporting evidence—"published efficient attention mechanism (NeurIPS 2024) subsequently deployed in production recommendation system handling 12M daily requests." If the posting says "ML Research Scientist," lead with publications, novel contributions, and benchmark results, then mention engineering skills as execution capability. ATS keyword profiles differ significantly between these roles—"model serving" and "Kubernetes" dominate engineering postings, while "novel architecture" and "state-of-the-art" dominate research postings ⁷.

Does the cloud platform I list matter for ATS ranking?

ATS matches the platform names present in the job posting. AWS SageMaker, Google Vertex AI, and Azure ML are three distinct keyword clusters—a resume listing only Azure experience will not match a posting searching for "SageMaker." If you have multi-cloud experience, list all platforms. If you have single-cloud experience, apply to postings that match your platform and consider earning a certification in a second cloud provider. AWS leads AI job postings at 32.9%, followed by Azure at 26% ⁵. Include both the service name and parent platform: "AWS SageMaker" rather than just "SageMaker" to ensure matching on both terms.

Should I include my GitHub profile and open-source contributions?

Include your GitHub URL in your contact header as plain text—ATS stores URLs as searchable strings but cannot crawl repositories. More importantly, translate your GitHub contributions into resume content. "Contributor to Hugging Face Transformers (3 merged PRs: optimized attention mask computation reducing memory allocation by 15%)" provides ATS keywords (Hugging Face, Transformers, attention mask, memory optimization) while demonstrating open-source engagement. Star counts and follower numbers are irrelevant to ATS but may catch a human reviewer's attention if notable (1,000+ stars on a personal project).

How should I present certifications versus a master's degree?

Both are ATS keywords, and both matter—but they signal differently. A master's degree in Computer Science, Machine Learning, or AI demonstrates foundational knowledge and research capability. Cloud certifications (AWS ML Engineer, Google Professional ML Engineer) demonstrate production deployment skills on specific platforms. List both. For entry-level candidates, the MS degree typically outweighs certifications. For mid-career and senior candidates, current certifications signal ongoing skill investment—Google and AWS ML certifications appeared in 40% more job postings than competing credentials ⁶. Expired certifications should be removed; they suggest lapsed skills.

What resume length is appropriate for AI engineers at different career stages?

One page for candidates with fewer than 3 years of experience and no publications. Two pages for engineers with 3+ years of production ML experience, published research, or significant open-source contributions. ATS does not penalize length, but human reviewers do—Jobscan data shows recruiters spend an average of 6-7 seconds on initial scan. A two-page resume for a junior engineer with one internship suggests poor editing. A one-page resume for a staff engineer with 9 years, 8 publications, and multi-team platform architecture suggests missing depth. If you have publications, include only the 3-5 most relevant rather than a full CV-style bibliography ⁴.

How do I optimize my resume when transitioning from data science to AI engineering?

Identify the overlapping keywords and lead with those: Python, model training, evaluation metrics, experiment tracking, SQL, feature engineering. Then add AI engineering-specific terms from the job posting: "model deployment," "inference optimization," "Docker," "Kubernetes," "API design," "latency," "throughput." Quantify any production-adjacent work from your data science role—dashboards serving 500 users, models running on scheduled batch pipelines, or A/B tests with statistical rigor. A strong transition resume reframes data science work through an engineering lens: "deployed XGBoost model to production via Flask API serving 2,000 daily predictions" rather than "built predictive model in Jupyter notebook."

References:

{
  "opening_hook": "The Bureau of Labor Statistics projects 20% employment growth for computer and information research scientists (SOC 15-1221) through 2034—nearly seven times the 3% average across all occupations—with a median annual wage of $140,910 and top earners exceeding $232,120. Meanwhile, AI-related job postings climbed from 1.4% to 1.8% of all U.S. job postings between 2023 and 2024 according to Stanford's AI Index Report, with Python appearing as the top specialized skill across those listings.",
  "key_takeaways": [
    "Framework-specific keywords determine ATS ranking—PyTorch appears in 37.7% of postings and TensorFlow in 32.9%, so listing 'deep learning frameworks' without naming them misses both matches",
    "Quantified model performance (inference latency, accuracy improvements, dataset sizes, GPU utilization) differentiates ranked resumes from rejected ones",
    "MLOps and deployment skills (Docker 15.4%, Kubernetes 17.6% of postings) are now table stakes for industry AI roles",
    "Cloud certifications (Google Professional ML Engineer, AWS ML) appeared in 40% more job postings than competing credentials",
    "Tables, two-column layouts, and graphics-based skill bars cause ATS parsers to scramble or drop critical technical content"
  ],
  "citations": [
    {"number": 1, "title": "Computer and Information Research Scientists - Occupational Outlook Handbook", "url": "https://www.bls.gov/ooh/computer-and-information-technology/computer-and-information-research-scientists.htm", "publisher": "Bureau of Labor Statistics"},
    {"number": 2, "title": "Occupational Employment and Wages - 15-1221 Computer and Information Research Scientists", "url": "https://www.bls.gov/oes/current/oes151221.htm", "publisher": "Bureau of Labor Statistics"},
    {"number": 3, "title": "Artificial Intelligence Index Report 2025", "url": "https://hai.stanford.edu/ai-index/2025", "publisher": "Stanford University HAI"},
    {"number": 4, "title": "The State of the Job Search in 2025", "url": "https://www.jobscan.co/state-of-the-job-search", "publisher": "Jobscan"},
    {"number": 5, "title": "AI Engineer Job Outlook 2025: Trends, Salaries, and Skills", "url": "https://365datascience.com/career-advice/career-guides/ai-engineer-job-outlook-2025/", "publisher": "365 Data Science"},
    {"number": 6, "title": "Top 10 AI Certifications Worth Getting in 2026 (ROI + Career Impact)", "url": "https://www.nucamp.co/blog/top-10-ai-certifications-worth-getting-in-2026-roi-career-impact", "publisher": "Nucamp"},
    {"number": 7, "title": "15-1221.00 - Computer and Information Research Scientists", "url": "https://www.onetonline.org/link/summary/15-1221.00", "publisher": "O*NET OnLine"},
    {"number": 8, "title": "The State of AI Hiring in 2025: Insights from 3,000 Job Listings", "url": "https://www.flex.ai/blog/the-state-of-ai-hiring-in-2025-insights-from-3-000-job-listings", "publisher": "Flex.ai"},
    {"number": 9, "title": "ATS Resume Formatting Research", "url": "https://www.topresume.com/career-advice/what-is-an-ats-resume", "publisher": "TopResume"}
  ],
  "word_count": 3250,
  "meta_description": "ATS optimization checklist for AI engineer resumes. Covers PyTorch, TensorFlow, LangChain keywords, MLOps deployment skills, model performance metrics, and format rules to pass automated screening for ML, NLP, and generative AI roles.",
  "prompt_version": "v2.0-cli"
}