ML Engineer Resume Example

A complete, annotated resume for a senior ML engineer building recommendation systems at scale. Every section is broken down — so you can see exactly what makes this resume land interviews.

Scroll down to see the full resume, then read why each section works.

Alex Petrov
alex.petrov@email.com | (917) 555-0381 | linkedin.com/in/alexpetrov | github.com/apetrov
Summary

ML engineer with 5+ years of experience building recommendation systems and real-time ML serving infrastructure at scale. Currently owning the candidate generation and ranking pipeline at Spotify, where I improved home feed CTR by 12% through a redesigned two-tower retrieval model serving 500M+ users. Strong track record of taking models from research prototype to production with rigorous A/B testing and monitoring.

Experience
Senior ML Engineer
Spotify New York, NY
  • Redesigned the candidate generation stage of the home feed recommendation pipeline, replacing a collaborative filtering model with a two-tower neural retrieval model that improved CTR by 12% and increased average session length by 8% across a 3-week A/B test on 5% of global traffic
  • Built a real-time feature serving layer using Feast and Redis, reducing feature freshness from 24 hours (batch) to under 30 seconds for user engagement signals, which enabled the ranking model to respond to within-session behavior changes
  • Reduced model training time from 14 hours to 3.5 hours by migrating the ranking model from single-node TensorFlow to distributed training on a 8-GPU Kubeflow pipeline with mixed-precision training and gradient accumulation
  • Led the design of the ML experiment framework used by 4 recommendation teams, standardizing A/B test analysis with pre-registered metrics, sequential testing for early stopping, and automated guardrail checks that prevented 2 regressions from shipping
ML Engineer
Wayfair Boston, MA
  • Owned the search ranking model for Wayfair’s product catalog (18M+ SKUs), improving NDCG@10 by 15% by replacing a gradient boosted tree model with a cross-encoder reranker on the top 100 candidates, after validating that the latency tradeoff (40ms additional p99) was acceptable for the quality gain
  • Built the feature engineering pipeline for search relevance using Spark and Airflow, processing 200M+ daily search events into 80+ features covering query-product similarity, user personalization, and business signals (margin, inventory, returns rate)
  • Designed and deployed a query understanding model that classified search intent (browse vs. specific product vs. attribute-based), improving null result rate by 30% by routing low-confidence queries to a broader retrieval strategy
  • Reduced model serving costs by 40% ($22K/month) by implementing model distillation, compressing the production ranking model from 350M parameters to 45M while retaining 97% of the original NDCG
Data Scientist
Capital One McLean, VA
  • Built credit risk models for the auto lending portfolio, developing a gradient boosted ensemble that improved Gini coefficient by 8% over the existing logistic regression baseline, resulting in an estimated $4.5M reduction in annual charge-offs
  • Developed an automated model monitoring system that tracked PSI (Population Stability Index) and feature drift across 15 production models, triggering retraining alerts that reduced model staleness incidents by 70%
Publications
  • Petrov, A. et al. “Efficient Two-Tower Models for Large-Scale Product Recommendation.” RecSys 2024. Described the retrieval architecture deployed at Wayfair, serving 50M monthly active users.
Skills

Languages: Python, Scala, SQL   ML Frameworks: PyTorch, TensorFlow, XGBoost, Hugging Face Transformers   Data & Features: Spark, Airflow, Feast, Kafka, BigQuery   Serving & Infra: Kubeflow, Docker, Kubernetes, TensorFlow Serving, Triton   Experiment: A/B testing frameworks, MLflow, Weights & Biases

Education
M.S. Statistics
Columbia University New York, NY

What makes this resume work

Seven things this ML engineer resume does that most don’t.

1

The summary is anchored in a specific ML domain

Not “ML engineer with experience in various machine learning projects.” Alex leads with recommendation systems and real-time ML serving — the exact domain and system type. A hiring manager for a recommendations team knows within five seconds that this person has directly relevant experience, not generic ML skills applied to an unrelated domain.

“...owning the candidate generation and ranking pipeline at Spotify, where I improved home feed CTR by 12% through a redesigned two-tower retrieval model serving 500M+ users.”
2

Bullets show the full ML lifecycle

Data pipelines (Spark processing 200M events), feature engineering (80+ features across three signal categories), model training (distributed training on GPU clusters), serving (real-time feature layer with 30-second freshness), and monitoring (PSI tracking, drift detection). This resume doesn’t just show someone who trains models — it shows someone who owns the entire system from data to production.

“Built the feature engineering pipeline for search relevance using Spark and Airflow, processing 200M+ daily search events into 80+ features...”
3

Model decisions are visible and justified

“Replacing a gradient boosted tree model with a cross-encoder reranker on the top 100 candidates, after validating that the latency tradeoff was acceptable.” This is the sentence that separates a senior ML engineer from everyone else. Alex doesn’t just pick a model — he evaluates the tradeoff between quality and latency, and tells you he validated it before shipping.

“...after validating that the latency tradeoff (40ms additional p99) was acceptable for the quality gain.”
4

A/B test results are quantified properly

Not “improved recommendations.” Alex specifies the metric (CTR), the magnitude (12%), the test duration (3 weeks), and the traffic allocation (5% of global traffic). This tells a hiring manager that Alex understands experiment design, not just model building. The guardrail checks and early stopping mention signals even deeper statistical rigor.

5

Infrastructure work sits alongside modeling

Many ML resumes are all models and no systems. Alex shows both: the two-tower retrieval model and the real-time feature serving layer, the ranking model improvement and the distributed training migration, the search relevance model and the feature engineering pipeline. This signals someone who can own the full stack, not just hand off a notebook to the platform team.

6

The publication is relevant and not padded

One publication at a top venue (RecSys), directly describing work that was deployed in production at Wayfair. No list of six co-authored workshop papers. No coursework projects disguised as research. The publication reinforces the production narrative rather than competing with it. If Alex had no publications, leaving this section out entirely would be fine — the experience speaks for itself.

“Described the retrieval architecture deployed at Wayfair, serving 50M monthly active users.”
7

Skills are organized by ML pipeline stage

Languages, ML Frameworks, Data & Features, Serving & Infra, Experiment — these categories mirror the stages of a production ML pipeline. A hiring manager can scan this in two seconds and confirm that Alex covers data processing, model training, model serving, and experiment analysis. This is more informative than a flat list and it shows a systems-level understanding of how ML work fits together.

Common resume mistakes vs. what this example does

Experience bullets

Weak
Trained machine learning models to improve product recommendations. Worked with the data team to analyze user behavior and optimize the recommendation engine.
Strong
Redesigned the candidate generation stage of the home feed recommendation pipeline, replacing a collaborative filtering model with a two-tower neural retrieval model that improved CTR by 12% and increased average session length by 8%.

The weak version describes a job function. The strong version names the specific pipeline stage, the architectural decision, the model type, and the measured business impact. Same work, completely different credibility.

Summary statement

Weak
Experienced ML expert with a strong background in machine learning, deep learning, and data science. Passionate about using AI to solve complex problems and drive business value.
Strong
ML engineer with 5+ years of experience building recommendation systems and real-time ML serving infrastructure at scale. Currently owning the candidate generation and ranking pipeline at Spotify, where I improved home feed CTR by 12%.

The weak version is a collection of adjectives. The strong version names the domain (recommendation systems), the scale (500M+ users), and a specific result — in two sentences that could only describe one person.

Skills section

Weak
Python, R, Java, Scala, TensorFlow, PyTorch, Keras, Scikit-learn, XGBoost, LightGBM, CatBoost, Spark, Hadoop, AWS, GCP, Azure, SQL, NoSQL, Deep Learning, NLP, Computer Vision, Recommender Systems
Strong
ML Frameworks: PyTorch, TensorFlow, XGBoost   Data & Features: Spark, Airflow, Feast, Kafka   Serving & Infra: Kubeflow, TensorFlow Serving, Triton   Experiment: MLflow, W&B

The weak version lists every ML framework ever made plus three cloud providers. The strong version is organized by pipeline stage and only includes tools actually used in production. It tells a coherent story about how Alex works, not just what Alex has heard of.

Frequently asked questions

How do I show ML research on an industry resume?
Only include research that’s relevant to the role you’re applying for, and frame it in terms of impact rather than methodology. Instead of listing every paper you’ve co-authored, pick the 1–2 that connect most directly to the job and describe them the same way you’d describe a work project: what problem it solved, what approach you took, and what the result was. If your paper led to a production system or influenced a product decision, say that. If you have significant publications, a single line like “Published 3 papers at NeurIPS/ICML on recommendation systems” is enough — link to your Google Scholar and move on.
Should I include Kaggle on my resume?
Only if you have genuinely strong results (top 1% in a relevant competition, Grandmaster status) and you’re early in your career. For experienced ML engineers with production work, Kaggle competitions can actually work against you — they signal competition-style ML (optimize a single metric on a fixed dataset) rather than production ML (build systems that serve, monitor, and improve models in the real world). If your best ML work is a Kaggle competition, that’s fine to include. But if you have production experience, lead with that and leave Kaggle off.
How do I quantify ML engineering work?
Use the metrics that ML teams actually track: model performance (AUC, precision/recall, NDCG for ranking), business impact (CTR lift, revenue per user, engagement), system performance (training time, inference latency, throughput), and cost (infrastructure spend, compute hours). The strongest bullets tie a technical improvement to a business outcome: “Redesigned the candidate generation model, improving CTR by 12% and driving an estimated $3.2M in annual incremental revenue.” If you can’t tie to revenue, tie to user metrics, engineering efficiency, or cost reduction.
1 in 2,000

This resume format gets you hired

This exact resume template helped our founder land a remote data scientist role — beating 2,000+ other applicants, with zero connections and zero referrals. Just a great resume, tailored to the job.

Try Turquoise free