Company interview guide

Grammarly (now Superhuman) Machine Learning Interview Guide (2026)

Practical ML coding, ML design and unusually deep career-history and values interviews.

Grammarly (now Superhuman)'s machine learning interview loop has 7 reported stages, from recruiter call to career history, values & hiring manager; typical length: 4-8 weeks, sometimes longer.

Roles covered: ML Engineer and Data Scientist. The question bank holds 1 question reported from Grammarly (now Superhuman) interviews.

Try a free question from this loop's topics

Now Superhuman. Grammarly bought the email app Superhuman in 2025 and renamed the company Superhuman in October 2025; Grammarly continues as a product.

Interview process length
4-8 weeks, sometimes longer
Stages
7
Reports linked to bank questions
1
Bank report dates
Nov 2017–Nov 2017
Questions in bank
1 (0 free)
Last reviewed
October 2026

How the Grammarly (now Superhuman) interview loop works

A recruiter call, sometimes followed by a short hiring-manager conversation, then a technical screen. For ML engineers the screen is usually one interview of coding plus ML concept questions (one report gives 75 minutes; another had two screens); interns may start with an online assessment and get plain coding or object-oriented design questions. The virtual onsite has about four to five rounds: practical ML coding (in some reports training a model in a hosted notebook), ML system design or ML theory, a research presentation or deep dive on a past project, and behavioral, hiring-manager and values interviews. Data science loops run longer, often six or more conversations, centered on SQL, product and business cases, experiment design and interpreting results; take-home analyses appear only in reports up to 2022. Reference checks after the onsite are mentioned in two reports. The rounds are well signposted: several candidates say the recruiter sent detailed preparation material for each one.

The Grammarly (now Superhuman) interview process, stage by stage

01Recruiter call

A phone or video call with a recruiter; some candidates then have a short call with the hiring manager

A conversation about your background, what you want next and why Grammarly. Several candidates say recruiters were unusually hands-on, sharing a guide for each later round and setting up extra chats with employees before a decision. A few data science candidates met the hiring manager next for about 30 minutes before any technical round.

What it tests

  • A clear summary of your background and goals
  • Motivation for the company and the specific team
  • Logistics: level, location, timeline

How to prepare

  • Ask for the preparation guide for every round; candidates say it describes each interview in useful detail.
  • Have a concise answer to why Grammarly that links its writing and AI products to your own work.
  • If you hold a statistics or research background, be ready to connect it explicitly to the role.

Recruiter or HR call in 4 ML-role and 9 data science reports, 6 since 2024; shared across tracks. The company's interviewing page lists it first, with an optional hiring-manager call.

02ML track: technical screen

Usually one live interview of coding plus ML concept questions (one report gives 75 min); some had two screens; some interns get an online assessment first

Candidates describe a medium-difficulty algorithmic problem together with short ML theory questions rather than hard puzzles. Intern and early-career candidates report plain coding questions on strings and stacks, or an object-oriented design task where you then speed up a method, for example by replacing a linear search with binary search.

What it tests

  • Writing correct, clean code for a medium algorithm problem
  • Short ML theory questions: models, losses, evaluation, overfitting
  • Object-oriented design and explaining your structure choices
  • Improving the time complexity of code you just wrote

How to prepare

  • Revise ML fundamentals out loud; the concept questions are quick and broad.
  • Practise string, stack and search problems at medium difficulty.
  • For object-oriented tasks, be ready to justify each class and discuss how you would extend it.

A coding or coding-plus-ML-concepts screen in 10 ML-role reports, 8 since 2024 (4 of them intern reports); one intern report starts with an online assessment. Details are thin in most.

03Data science track: technical screen or take-home

One or more live SQL and product-case screens; older reports had a take-home analysis in a notebook instead

Recent data science candidates describe live SQL written against a described schema with no data to run it on, joined with a business or product case. Senior candidates had several screens, including reading charts and interpreting experiment results. Up to 2022 several candidates instead got a take-home on app usage pings (descriptive stats, retention curves, which referral sources bring the best users); a 2024 candidate notes there was no take-home.

What it tests

  • Multi-table SQL joins and aggregations without running the query
  • Turning a vague product question into metrics
  • Reading charts and experiment results correctly
  • Retention and cohort analysis from raw event logs

How to prepare

  • Practise writing SQL on paper against a schema you only hear described.
  • If you get a take-home, scope it: answer each question directly before adding extras, since several candidates spent many hours and got no feedback.

Live SQL or case screens in 6 data science reports (3 since 2024) and a take-home in 4 reports from 2017-2022, two on usage-ping data.

04ML track: onsite technical rounds

Part of a virtual onsite of about 4-5 rounds: practical ML coding, ML system design and in some loops ML theory or a large-scale data round

The technical half of the onsite tests hands-on work more than algorithm drills. Candidates mention ML coding focused on practical, everyday tasks, one round that meant training a model end to end in a hosted notebook, an ML system design interview, and in a 2023 loop a separate ML theory round and a big-data computation round.

What it tests

  • Building and training a model end to end in a notebook
  • Practical day-to-day ML code: data handling, features, evaluation
  • ML system design for a product feature
  • ML theory depth beyond the screen
  • Reasoning about computation over large datasets

How to prepare

  • Rehearse a full workflow in a hosted notebook: load data, baseline, train, evaluate, explain.
  • Prepare an ML design for a writing-assistance feature such as grammar correction or tone suggestions, with offline and online metrics.
  • Ask the recruiter whether your loop includes theory or a data-processing round; formats differ by team.

ML coding, ML design or ML theory onsite rounds in 5 ML-role reports, only 2 since 2024. The company's interviewing page describes technical sessions with team members over several hours.

05Data science track: onsite rounds

A virtual onsite of roughly 5-7 one-to-one conversations, 45-60 min each: SQL and case, A/B testing, product sense (sometimes with a PM), plus behavioral

Candidates describe practical rounds with no trick questions: SQL with a case study, designing and analysing an experiment for a new feature, product knowledge, and reading graphs. Some loops add statistics, Python and ML concept questions. Data science loops are long, and one candidate counted more than ten conversations over several months.

What it tests

  • Designing an experiment to judge whether a feature met its business goal
  • Interpreting experiment results and charts
  • SQL and data manipulation tied to a business question
  • Product sense for a writing assistant
  • Statistics and basic ML concepts

How to prepare

  • Practise A/B test design end to end: metric, unit of randomization, duration, guardrails and how you would read a null result.
  • Use the product before the loop and form views on what success looks like for its features.
  • Plan your energy; candidates mention six to seven or more sessions.

Onsite rounds in 10 data science reports, 6 since 2024; SQL, product cases and experiments recur. The company's interviewing page describes the multi-session technical round.

06Research presentation & project deep dive

A presentation of past work or a detailed discussion of one or two projects; some loops add extra deep-dive interviews after the main onsite

ML candidates, especially those with a research background, are often asked to present past work and then field detailed questions on choices and results. Others had a round devoted to one previous ML project, or were given two extra interviews on their area of expertise after passing the onsite.

What it tests

  • Explaining a project's goal, approach and measured impact
  • Defending modelling and evaluation choices
  • Depth in your stated area of expertise
  • Honesty about what you personally did

How to prepare

  • Prepare a short talk on one project with a clear problem, method, results and what you would change.
  • Expect follow-ups on data, baselines and failure cases rather than only the headline result.

Research presentation or project deep dive in 6 ML-role reports, 3 since 2024 (two give a research presentation).

07Career history, values & hiring manager

Two or more conversations, 30-60 min, with the hiring manager, someone from outside the hiring team and future teammates

The most distinctive part of the loop. Several candidates had one or two interviews that walk through every past job in order, asking about successes, failures, managers and what each step taught them. Behavioral answers are judged against the company values, and one interview is often with someone outside your hiring org.

What it tests

  • A coherent story across each job you have held
  • Self-awareness: strengths, growth areas, lessons from failures
  • Collaboration and handling conflict, with specific examples
  • Fit with the company values

How to prepare

  • Write a short note for each past role: why you joined, what you shipped, what went wrong, why you left.
  • Use specific situation, action and result examples; read the current values page, which the company revised after the rename.
  • Keep references from managers and peers ready; two candidates were asked for them after the onsite.

Behavioral, values, career-history or hiring-manager rounds in 7 ML-role and 8 data science reports, 7 since 2024. The company's interviewing page describes values-based behavioral and manager interviews.

Questions and themes reported most often

  • Present or deep-dive a past ML or research project (6 reports, research presentation & project deep dive)
  • Design and analyse an experiment for a new product feature (4 reports, data science track: onsite rounds)
  • SQL joins across several tables without data to run on (4 reports, data science track: technical screen or take-home)
  • Walk through your career job by job: successes, failures and lessons (3 reports, career history, values & hiring manager)
  • Medium coding problem plus short ML theory questions (3 reports, ml track: technical screen)
  • ML system design interview (3 reports, ml track: onsite technical rounds)
  • Hands-on ML coding, such as training a model in a hosted notebook (3 reports, ml track: onsite technical rounds)
  • Take-home on app usage pings: retention curves and best referral sources (2 reports, data science track: technical screen or take-home)

What Grammarly (now Superhuman) asks about

  • Behavioral, values & career history25%
  • Product case & experimentation18%
  • General coding15%
  • SQL & data analysis15%
  • ML coding & practical modelling14%
  • ML system design & theory13%

Grammarly (now Superhuman) interview prep checklist

  • Wrote a one-paragraph story for every past role, including a failure and a lesson
  • Prepared a short presentation on one ML project with measured results
  • Trained and evaluated a model end to end in a notebook within an hour
  • Sketched an ML design for a writing-assistance feature with offline and online metrics
  • Solved several medium string, stack and search problems cleanly
  • Data science: designed an A/B test for a new feature and practised SQL joins on paper
  • Read the company's current values page and matched a story to each value

Tips

  • Prepare your career narrative as seriously as the technical rounds; whole interviews are spent on it.
  • Ask your recruiter for the round-by-round guide and use it to decide what to drill.
  • For ML roles, practise training and evaluating a model in a hosted notebook while explaining each step.
  • Expect practical, everyday coding rather than hard algorithm puzzles, but keep medium-level problem solving sharp.
  • Have one project ready as a short presentation with results, trade-offs and failures.
  • For data science roles, drill SQL without running it, product cases and experiment design.
  • Think about ML for writing assistance: grammar correction, rewriting, tone, and how to evaluate them.
  • The loop predates the October 2025 rename in all our reports, so confirm the current rounds before you prepare.

Practice the Grammarly (now Superhuman) loop

Work the stages in order: every reported question for this company, grouped by round, with solutions in the app.

Grammarly (now Superhuman) interview FAQ

How many interview rounds does Grammarly (now Superhuman) have for ML roles?

Candidates report 7 stages: Recruiter call, ML track: technical screen, Data science track: technical screen or take-home, ML track: onsite technical rounds, Data science track: onsite rounds, Research presentation & project deep dive and Career history, values & hiring manager. A recruiter call, sometimes followed by a short hiring-manager conversation, then a technical screen. For ML engineers the screen is usually one interview of coding plus ML concept questions (one report gives 75 minutes; another had two screens); interns may start with an online assessment and get plain coding or object-oriented design questions. The virtual onsite has about four to five rounds: practical ML coding (in some reports training a model in a hosted notebook), ML system design or ML theory, a research presentation or deep dive on a past project, and behavioral, hiring-manager and values interviews. Data science loops run longer, often six or more conversations, centered on SQL, product and business cases, experiment design and interpreting results; take-home analyses appear only in reports up to 2022. Reference checks after the onsite are mentioned in two reports. The rounds are well signposted: several candidates say the recruiter sent detailed preparation material for each one.

How long does the Grammarly (now Superhuman) ML interview process take?

Reported timelines run 4-8 weeks, sometimes longer.

What does Grammarly (now Superhuman) ask in ML interviews?

By reported question mix: Behavioral, values & career history 25%; Product case & experimentation 18%; General coding 15%; SQL & data analysis 15%; ML coding & practical modelling 14%; ML system design & theory 13%. The most repeated themes: Present or deep-dive a past ML or research project; Design and analyse an experiment for a new product feature; SQL joins across several tables without data to run on.

How should I prepare for the Grammarly (now Superhuman) ML interview?

Prepare your career narrative as seriously as the technical rounds; whole interviews are spent on it. Ask your recruiter for the round-by-round guide and use it to decide what to drill. For ML roles, practise training and evaluating a model in a hosted notebook while explaining each step.

Are there free Grammarly (now Superhuman) interview questions?

LeetML has 1 question reported from Grammarly (now Superhuman) interviews. The stage sections above link the free practice problems for the topics each round covers; the reported-question set itself is in the app.

Related interview guides

Official sources