Harvard · Resume

Data · 2026

Harvard Resume for Data Scientists & ML Engineers

Data science hiring splits into two tracks: applied (drive a business metric) and research (publish, train models, contribute to OSS). The Harvard format works for both. This recipe shows how to pivot the same skeleton — Education first, Experience with quantified impact, projects/publications — toward either track.

Start composingfree · no signup
Harvard Resume··~5 min

What recruiters look for

  • Modelling depth (architectures, frameworks, eval methodology)
  • Causal/inferential rigor (A/B test design, confounders handled)
  • Productionisation experience (not just notebook output)
  • Publications, OSS contributions, Kaggle medals (for research-track)
  • Business outcomes attributable to your models (for applied-track)

Required sections, in this order

Track selection

  • Applied track: lead Experience with business-metric bullets; Publications optional
  • Research track: surface Publications section between Education and Experience; longer Skills section with frameworks listed

Skills section content

  • Languages: Python, R, SQL, Scala (only what you're conversant in)
  • ML frameworks: PyTorch, JAX, scikit-learn, XGBoost
  • Infrastructure: Spark, Airflow, dbt, Databricks, Snowflake
  • Cloud + MLOps: AWS/GCP, MLflow, Weights & Biases, BentoML

Sample in Harvard format

Harvard Resume for Data Scientists · 2026 Template & Guide
Harvard format · 1 page

Strong vs weak bullets

Before

Built a recommendation model for the marketplace

After

Built and shipped a two-tower recommendation model (PyTorch, BigQuery embeddings) replacing a collaborative-filtering baseline; A/B tested over 6 weeks across 8M users; +14.2% click-through and +$3.7M monthly GMV

Architecture (two-tower), data infra (BigQuery embeddings), what it replaced, A/B duration + sample, and dual metrics (CTR + GMV) — a senior reviewer infers full-cycle ownership.

Before

Authored a paper on transformer efficiency

After

Co-authored 'Efficient Sparse Attention for Long-Context Transformers' (NeurIPS 2025 main); reduced inference cost 38% at comparable accuracy on 4 standard benchmarks; cited by 12 papers in first 6 months

Venue (NeurIPS main), specific contribution (sparse attention), measurable improvement (-38% cost), and downstream impact (citations).

Before

Improved A/B testing infrastructure

After

Redesigned the experimentation platform's sequential testing module to support 4 simultaneous treatments per surface; cut median experiment duration from 28 to 11 days; adopted by 12 product pods

Names what changed (sequential testing for multi-treatment), measurable speed (28 → 11 days), and adoption (12 pods).

Mistakes specific to this role

  • Listing every ML algorithm from a textbook. Pick 5-8 you've actually deployed.
  • Omitting business outcomes for applied roles. Models don't matter without a metric they moved.
  • Hiding Kaggle medals — if you have a gold or 2+ silvers, surface them under Awards.
  • Overweighting an undergrad coursework section over your real Experience. Coursework is for new grads only.

Your résumé starts here. Pay later.

Start composing

Frequently asked

Should I include LeetCode for DS roles?
Most DS roles don't filter on LeetCode the way engineering roles do. Top tier (Google Research, Anthropic) may quiz, but it's not a résumé filter. Skip unless your rating is top 1%.
Where do Kaggle competitions go?
Under Awards if you medalled in a Featured competition, under Projects otherwise. Don't list every competition you entered.
How do I show I can productionise, not just prototype?
Lead at least one bullet per role with a deploy verb (shipped, rolled out, productionised) + a uptime/latency/throughput metric.

Related