Company interview guide

Glean Machine Learning Interview Guide (2026)

Graph-search coding, an AI-assisted build round and enterprise ML design, judged on a small set of reports.

Glean's machine learning interview loop has 7 reported stages, from recruiter or hiring-manager intro to data scientist track: sql, statistics, take-home, case.

Roles covered: ML Engineer, AI Engineer and Data Scientist. The question bank holds 16 questions reported from Glean interviews, 2 of them free to solve in your browser.

Try a free question

Stages
7
Reports linked to bank questions
19
Bank report dates
Jan 2025–Sep 2026
Questions in bank
16 (2 free)
Last reviewed
October 2026

How the Glean interview loop works

Reports describe a recruiter or hiring-manager intro, one technical phone screen with an algorithmic problem, then an onsite (in person in San Francisco in a 2026 report) with ML design, an AI-assisted build and a coding round. Some candidates met the hiring manager first with no recruiter call, and one 2025 report had two coding rounds, a resume discussion, ML design and a hiring-manager round. Few reports describe a behavioral round or the decision step, and two candidates heard little or nothing after their final rounds. Because the outline rests on few reports, ask which rounds you will get.

The Glean interview process, stage by stage

01Recruiter or hiring-manager intro

Short call; some candidates go straight to the hiring manager

Most candidates start with a recruiter, who in one report also said the screen would be standard data structures and algorithms. At least one ML candidate was booked directly with the hiring manager without a recruiter call, so be ready to discuss your projects from the first conversation.

What it tests

  • Why Glean and why enterprise search or assistants
  • How your ML, search or LLM work maps to the team
  • Logistics: level, location and the rounds ahead

How to prepare

  • Treat any first call as a possible hiring-manager conversation and have a two-minute project summary ready.
  • Ask for the type and length of each later round, including whether an AI coding assistant is required.
  • ML job postings ask which AI tools you use and how; have a concrete answer.

Two ML reports describe a recruiter screen, one ML report a hiring-manager call in place of it, and one data science report a first screening round; formats and lengths are not reported.

02Technical phone screen

One live coding session, about 60 min in public guides; algorithmic problem with several parts

The ML screen is a general coding round rather than an ML quiz. The train-schedule reachability problem appears in a 2025 screen report and again in 2026 onsite reports. Earlier reports describe problems that are simple algorithmically but need a lot of input parsing and careful code.

What it tests

  • Modelling a problem as a graph over (place, time) states and running BFS
  • Handling waiting, transfers and multi-part follow-ups
  • Writing complete, working code that passes the interviewer's tests
  • Parsing messy string input without losing time

How to prepare

  • Practise BFS over composite states such as (station, time) and how to rebuild the path from parent pointers.
  • Talk through your plan even if the interviewer is quiet; one candidate passed every test case and was still rejected.
  • Budget time for input parsing and edge cases, not just the core algorithm.

Four ML reports from 2024-2026 describe a coding screen before the onsite (one names only the round); a 2021 report is not counted. Public guides add one problem of about 60 minutes.

03Onsite: ML system design

About 60 min, open-ended discussion

The repeated 2026 prompt is an employee-to-employee distance (or similarity) system that powers people search, networking and expert discovery inside a company. Expect to define what distance means for each use, choose signals, learn and evaluate it, and serve it for every pair. Compiled public lists add permission-aware enterprise RAG and ranking for a tenant with no interaction data.

What it tests

  • Framing a vague product goal as a measurable ML objective
  • Choosing signals such as org chart, documents, meetings and messages, and respecting privacy and permissions
  • Learning and evaluating a similarity or ranking model without clean labels
  • Serving pairwise scores efficiently at company scale
  • Cold start for new employees or new customers

How to prepare

  • Spend the first minutes on what the distance is for; different products want different definitions.
  • Bring up permissions and privacy early; enterprise data access is central to the product.
  • Name concrete offline and online metrics and say how you would get labels.

Three ML reports (2025, two in 2026) describe an ML design round; both 2026 reports give the employee-distance prompt. Enterprise RAG and ranking prompts come from a compiled public list.

04Onsite: AI-assisted build

75 min; you must use an AI coding assistant such as Cursor or Claude Code

A newer round in which you build a working program with an AI assistant and are expected to understand, change and explain its code. The reported task is a rate-limited Wikipedia crawler that first visits pages whose titles start with letters not yet seen, then crawls freely once all 26 letters are covered.

What it tests

  • Planning the work before prompting, then steering the assistant
  • Reading and fixing generated code quickly
  • Rate limiting, a visited set and a priority rule for the crawl frontier
  • Explaining design choices in code you did not type yourself

How to prepare

  • Practise with the assistant you will use until you can plan, prompt and review in short loops.
  • Write the core data structures (frontier, visited set, letter coverage) yourself or spell them out precisely; one candidate lost time decoding generated code.
  • Keep the program runnable at every step and test against a small page set.

Two 2026 ML onsite reports describe a 75-minute assisted crawler build. Glean's own ML, applied scientist and data scientist postings say the process includes a brief AI-focused exercise or discussion, without giving its length.

05Onsite: coding

About 60 min live coding; one 2025 loop had two coding rounds

A graph or grid problem, often with a second part. In 2026 this was the train-timetable problem again: BFS over (station, time) states where you may stay on a train or change, with a follow-up on printing the full route. The interviewer said finishing both parts was not required.

What it tests

  • Earliest arrival and reachability over a time-expanded graph
  • Reconstructing the actual path, not just its length
  • Shortest paths on grids with obstacles
  • Clean code under time pressure

How to prepare

  • Get part one fully working before discussing the follow-up.
  • Know BFS path reconstruction cold; it came up as a follow-up in more than one report.
  • Grid shortest path with the actual route returned is worth a dedicated drill.

Two 2026 ML onsite reports describe a 60-minute train-timetable coding round; a 2025 ML report describes two harder coding rounds including grid shortest path.

06Resume deep dive and hiring manager

Conversation about past projects; length not reported

A resume discussion and a hiring-manager round appear in one 2025 ML loop. Expect detailed questions about retrieval, ranking or LLM systems you have shipped. The specific retrieval topics listed here, such as chunking, hybrid search and RAG versus fine-tuning, come from a compiled public list, not from candidates.

What it tests

  • Depth on a search, ranking or LLM system you built
  • Retrieval choices: chunking, sparse plus dense search, reranking
  • When RAG beats fine-tuning for company knowledge, and where it fails
  • Latency trade-offs across retrieve, rerank and generate

How to prepare

  • Pick one project and prepare its metrics, failures and what you would change.
  • Be ready to explain hybrid retrieval and how you would measure retrieval quality.
  • Connect your answers to shipping reliable systems; the job postings stress production work over prototypes.

Free reported questions for this stage

One 2025 ML report lists a resume discussion and a hiring-manager round. Topic lists come from a compiled public write-up; no first-hand ML report describes a behavioral round.

07Data scientist track: SQL, statistics, take-home, case

Several rounds: SQL, statistics with derivations, a multi-part Python take-home, a product case and a product leader

Data scientist candidates describe a longer loop than ML engineers. One had a ten-part data exploration take-home due over a weekend; the statistics round was a probability quiz on expected values. Another went through SQL, statistics, a take-home, an analytics case and a head-of-product round.

What it tests

  • SQL for product metrics
  • Hypothesis tests such as a z-test and product KPIs
  • Expected values of statistics of random draws, derived by hand
  • Exploratory analysis and clear write-up in a take-home

How to prepare

  • Ask exactly what each round covers; one candidate was told to expect SQL in a statistics round that had none.
  • Plan your calendar for a short take-home deadline.
  • Drill order statistics and expectations of minima and maxima of uniform draws.

Two 2024 data science reports (one senior, one in Bengaluru) agree on SQL, statistics and a take-home; only one mentions a case study and a product-leader round.

Questions and themes reported most often

  • Train schedule reachability with waiting and transfers, then print the route (3 reports, technical phone screen)
  • Design an employee-to-employee distance for people search and expert discovery (2 reports, onsite: ml system design)
  • Rate-limited Wikipedia crawler with first-letter coverage, built with an AI assistant (2 reports, onsite: ai-assisted build)
  • Shortest path in a grid, returning the actual path (1 reports, onsite: coding)
  • Minimum distance to goal posts on a grid with heavy input parsing (1 reports, technical phone screen)
  • Group words that are synonyms in a list of strings (1 reports, technical phone screen)
  • Expected values of statistics of a random pair of numbers (1 reports, data scientist track: sql, statistics, take-home, case)
  • Product KPIs, a z-test and SQL questions (1 reports, data scientist track: sql, statistics, take-home, case)

What Glean asks about

  • Coding (graphs, BFS, parsing)28%
  • ML / search system design27%
  • AI-assisted build15%
  • Retrieval & LLM concepts15%
  • Resume & hiring manager8%
  • Stats & SQL (data science)7%

Glean interview prep checklist

  • Solved train-schedule reachability and earliest arrival with BFS, and printed the route
  • Built a rate-limited crawler with a priority frontier in 75 minutes using an AI assistant
  • Can design an employee-distance service from goal and signals through evaluation and serving
  • Can sketch permission-aware retrieval for an enterprise assistant
  • Returned the actual path for a grid shortest-path problem without bugs
  • Prepared a deep dive on one shipped search, ranking or LLM project

Tips

  • Drill BFS over time-expanded graphs until (station, time) states and path reconstruction feel routine; the train problem is the most repeated question.
  • Rehearse a full 75-minute build with your AI coding assistant, including planning, reviewing and explaining its code.
  • Prepare an ML design answer for similarity between people or documents inside a company, with permissions and privacy built in.
  • Know enterprise retrieval basics: hybrid sparse and dense search, chunking, reranking and evaluation without customer data.
  • Narrate your thinking even when the interviewer says little; passing tests alone did not secure an offer for one candidate.
  • Confirm each round's content in writing; candidates report surprises in round descriptions.
  • Follow up politely if you hear nothing after the final rounds; slow or missing decisions are reported.

See the rest of the Glean loop

The later stages, the questions reported for them and how to prepare are in Pro.

Unlock with Pro

Practice the Glean loop

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

Glean interview FAQ

How many interview rounds does Glean have for ML roles?

Candidates report 7 stages: Recruiter or hiring-manager intro, Technical phone screen, Onsite: ML system design, Onsite: AI-assisted build, Onsite: coding, Resume deep dive and hiring manager and Data scientist track: SQL, statistics, take-home, case. Reports describe a recruiter or hiring-manager intro, one technical phone screen with an algorithmic problem, then an onsite (in person in San Francisco in a 2026 report) with ML design, an AI-assisted build and a coding round. Some candidates met the hiring manager first with no recruiter call, and one 2025 report had two coding rounds, a resume discussion, ML design and a hiring-manager round. Few reports describe a behavioral round or the decision step, and two candidates heard little or nothing after their final rounds. Because the outline rests on few reports, ask which rounds you will get.

What does Glean ask in ML interviews?

By reported question mix: Coding (graphs, BFS, parsing) 28%; ML / search system design 27%; AI-assisted build 15%; Retrieval & LLM concepts 15%; Resume & hiring manager 8%; Stats & SQL (data science) 7%. The most repeated themes: Train schedule reachability with waiting and transfers, then print the route; Design an employee-to-employee distance for people search and expert discovery; Rate-limited Wikipedia crawler with first-letter coverage, built with an AI assistant.

How should I prepare for the Glean ML interview?

Drill BFS over time-expanded graphs until (station, time) states and path reconstruction feel routine; the train problem is the most repeated question. Rehearse a full 75-minute build with your AI coding assistant, including planning, reviewing and explaining its code. Prepare an ML design answer for similarity between people or documents inside a company, with permissions and privacy built in.

Are there free Glean interview questions?

Yes. LeetML has 16 questions reported from Glean interviews, and 2 of them are free to solve in the browser; they are listed by stage above.

Related interview guides

Official sources