Machine Learning CV Example

Format and length

Keep your machine learning CV to one page if you have 0-5 years of experience, two pages for senior roles. Use reverse-chronological order because in ML the recency of your stack is itself a signal. Your most recent work shows whether you are current on PyTorch, Kubeflow, LLMs, and modern serving tools.

Section order

1. Contact details and personal statement at the top (name, location, email, phone, LinkedIn).
2. Skills grouped into named families (Core ML, Deep Learning Frameworks, MLOps & Serving, LLM & GenAI, Cloud & Infrastructure, Data Pipelines, Monitoring). ATS and recruiters scan faster for Kubeflow or Triton under an MLOps heading than buried in a flat list.
3. Experience in reverse-chronological order, with 3-4 achievement bullets per role. Each bullet pairs a technical choice with a production outcome and a number.
4. Education (degree, institution, dates, key modules or dissertation if relevant).
5. Achievements (certifications, Kaggle medals, publications, open-source contributions).
6. Additional information (languages, interests) only if genuinely relevant.

Personal statement

Two to three sentences: your ML sub-specialty (NLP, computer vision, recommender systems), years of experience, one standout production metric, and your core technical stack. Anchor to a clear title (ML Engineer, AI Engineer, Data Scientist, ML Research Scientist) because each routes you to different interview pipelines and salary bands. If your work is 70% or more training and deploying models, claim ML Engineer, not Data Scientist.

Experience bullets

Every bullet follows the same pattern: technical choice plus production outcome plus number. Weak bullets list duties. Strong bullets show what you built, how you built it, and what changed. Always include the model architecture or tooling decision, the deployment target, and a metric (AUC lift, p95 latency, cost per inference, uptime, business impact).

Skills

Group your skills into named families with clear headings. List PyTorch before TensorFlow (PyTorch appears in 37.7% of ML postings and dominates 85% of research). Only list MLOps and serving tools you have actually shipped to production. Owning one full stack (Kubeflow + MLflow + Feast + Triton) outweighs name-dropping ten tools you only touched in tutorials. Add an LLM & GenAI heading if you have real experience with RAG, LoRA fine-tuning, vLLM, or vector databases.

Education and certifications

List your degree, institution, and dates. Include your dissertation topic if it is relevant to the role. Add key modules if you are a recent graduate. Certifications (AWS ML Specialty, GCP ML Engineer) go in the Achievements section, not Education.

What to leave out

No photo, no date of birth, no marital status (UK CV convention). Do not list every online course. Do not claim Kaggle Expert without naming competitions and ranks. Do not put BLEU scores on a recommender-system bullet or MAP@K on an NLP project. Match the performance metric to your sub-specialty or you signal you do not know the domain.

The result-plus-metric pattern

Every ML bullet should answer three questions: what did you build, how did you build it, and what changed. The most common failure mode is an ML CV that reads like a research paper instead of a production system. Recruiters separate ML engineers from data scientists by deployment-focused language. Show that you shipped, served, and monitored models, not just trained them in a notebook.

Weak vs strong bullets

Match metrics to your sub-specialty

• Computer Vision: mAP, IoU, inference time (ms), FPS
• NLP: BLEU, ROUGE, perplexity, F1, accuracy
• Recommender systems: MAP@K, NDCG, CTR, conversion lift
• General classification: accuracy, precision, recall, F1, AUC-ROC

Putting BLEU on a recommender-system bullet or MAP@K on an NLP project signals you do not know the domain.

When metrics are NDA-protected

Use relative improvements and infrastructure outcomes instead of raw confidential numbers: improved AUC by 12%, ingested 4.3B events per day, served 31 downstream models at p99 freshness of 47s. Never fabricate exact benchmarks. Scale and relative lift are credible without disclosing proprietary data.

Action verbs for ML roles

• Model development: trained, fine-tuned, architected, designed, implemented, optimised
• Deployment: deployed, served, converted, migrated, containerised, scaled
• Infrastructure: built, automated, orchestrated, integrated, configured
• Impact: improved, reduced, increased, achieved, maintained, drove
• Leadership: led, mentored, established, authored, presented

Education

List your degree, institution, field of study, and dates (start year and end year). If you graduated with honours (First Class, Upper Second), include it. For recent graduates, add your dissertation or final-year project title if it is relevant to ML roles, along with a one-line summary of the technique and result.

Include key modules only if you are within two years of graduation and the modules are directly relevant (Deep Learning, Computer Vision, NLP, Reinforcement Learning). After two years, your production experience carries more weight than your coursework.

Certifications that matter

ML certifications signal you understand cloud ML tooling and production workflows. The two that carry the most weight are:

• AWS Certified Machine Learning – Specialty: covers SageMaker, data pipelines, model deployment, and monitoring.
• Google Cloud Professional Machine Learning Engineer: covers Vertex AI, TensorFlow, model serving, and MLOps.

List certifications in the Achievements section with the issuing body and year. Do not list every online course or unverified certificate. Coursera and Udemy certificates without proctored exams do not move the needle.

PhDs and research experience

If you have a PhD, list it in the Education section with your thesis title and a one-line summary. Include your supervisor if they are well known in the ML community. For applied ML engineering roles, a PhD is valuable but not dominant. A shipped model with measurable business impact is the single highest-weight signal. Non-PhDs routinely beat PhDs for these roles by leading with deployed, monitored, revenue-moving work.

A PhD becomes the dominant signal only for dedicated ML Research Scientist tracks, where publications, benchmark wins, and novel contributions are the primary evaluation criteria.

Publications and open source

List publications in the Additional Information section with full citations in plain text format: Author, A., Author, B. (Year). Title of the paper. Conference/Journal Name, volume(issue), pages. DOI or ArXiv link. Venue beats count. NeurIPS, ICML, ICLR, ACL, EMNLP, and CVPR carry real weight. First-author beats co-author. Bare ArXiv preprints without peer review do not move the needle.

For open-source contributions, list only merged PRs to well-known projects (PyTorch, Hugging Face, vLLM, TensorFlow) with the PR number and a one-line impact. Maintaining your own library counts if it has meaningful adoption (100+ stars, used in production by other teams).