Glean interview, decoded.

Candidate reports describe four to five stages over roughly three to five weeks. The phone screen problems below come from reports dated November 2025 to August 2026; the shape of the onsite comes from third-party guides and candidate snippets, so ask the recruiter for your exact rounds.

The ads and the reports point the same way: ship working things, stay close to the customer, and take whatever work matters most.

• Working code over a polished fragment, in the build round most of all
• Customer focus: engineers talk to customers and work from their problems
• Ownership of a feature from idea to launch and after
• Taking on whatever is most useful to the company, a line in every engineering ad
• Fluency with AI tools in your own work

The coding problems are practical and recur across reports, so they have their own section below. The behavioural questions follow the values in the ads.

These are the problems candidates have named for Glean's phone screen and onsite between November 2025 and August 2026, on 1Point3Acres and PracHub. The wording changes by interviewer, so what follows is each problem and the approach that solves it. Plan to write it so it runs, with a quick test at the end.

• Implement a byte pair encoding (BPE) tokenizer: Start from characters. Count every adjacent pair across the corpus, merge the most frequent pair into a new token, record the merge, and repeat for a fixed number of merges. To encode new text, apply the recorded merges in the order they were learned. Be ready to talk about tie-breaking and the cost of recounting after each merge. Reported twice in the phone screen, June 2026.
• Word search in a grid: Depth-first search from each cell, marking cells as used on the way down and unmarking on the way back. If you are given many words, build a trie and walk it during the search so you stop early on dead prefixes.
• Find the kth largest across two sorted arrays: The linear answer is a two-pointer merge. The expected follow-up is logarithmic: compare the element k/2 in from each array and discard the half that cannot contain the answer, shrinking k each time.
• 2048: implement the tilt move: Solve one row moving left: drop the zeros, merge equal neighbours once from the front, pad with zeros. Every other direction is the same function on a reversed row or a transposed board. Reported in February and March 2026.
• Simulate assigning documents to indexers: A queue of documents and m indexers working in parallel. Keep a min-heap of indexers keyed by when each becomes free; pop the earliest, assign the next document, push it back with its new finish time. State the tie-break rule before coding.
• Object-oriented design: chess: A board, a piece type per kind with its own move rules, a game that tracks turns and checks legality. Keep move validation on the pieces and the game state in one place, and say what you would add next (check, castling, en passant).
• Restore an array with one displaced element: One element was moved out of a sorted array. Find where the order breaks in one pass, take the displaced element out, and reinsert it where it belongs, in linear time.
• Return top department suggestions for a prefix: A typeahead: match the prefix with a trie or binary search over sorted names, then return the top k by the ranking you are given, using a heap if the list is long.

Candidates also describe a two-hour practical build at the onsite, in their own IDE. Pick a structure in the first ten minutes, get the simplest version running end to end, then improve it. The reported failure is spending the first hour planning.

Levels.fyi figures for Glean software engineers, read on 23 September 2026: a median package of $222K across the US and $330K in the San Francisco Bay Area, made up of base, stock options and bonus. Glean posts base ranges on every US engineering ad: $140K to $265K for fullstack, storage and search quality, $175K to $270K for AI and ML infrastructure, and $210K to $300K for the APIs and billing platform teams, plus equity. Bangalore roles post no range. Calibrd benchmarks the pay on any posting you scan against its level and location.

1. 01Write the BPE tokenizer from memory once, then time yourself on 2048's tilt move and word search. These three come up more than anything else.
2. 02Practise building something that runs in two hours: a small service or a data task with a real interface. Get a first version working in the first twenty minutes.
3. 03Prepare a search design: connectors pulling from many apps, an index kept fresh, ranking, and results filtered by what each person is allowed to see.
4. 04Have a real answer to how you use AI in your work, with an example. Every ad says there is an exercise or discussion on it.
5. 05Check the office days for the team before the loop. US roles are hybrid, three or four days a week.

This guide covers Glean's software and machine learning engineering loops. Sales, solutions and customer engineering roles run a different process. For management and leadership roles the loop is similar but the bar shifts to people, delivery and strategy, so pair it with the leadership interview prep hub. The bar for your exact role comes from the role-by-role guides, and the prep that actually transfers is spoken, so run a mock interview before the real one.

• 011Point3Acres, Glean interview reports21 reports from November 2025 to August 2026, most of them phone screens, naming the BPE tokenizer (June 2026), OOD chess (June 2026), word search (April 2026), kth largest from two sorted arrays (April 2026), a processor problem (January 2026) and a document indexing challenge (November 2025)
• 02PracHub, Glean software engineer questionsthe BPE tokenizer (June 2026), kth largest in two sorted arrays, word search in a grid, 2048 game logic and the tilt move (February and March 2026), restoring an array with one displaced element, top department suggestions, and simulating document assignment to indexers
• 03Blind, Glean assignment coding roundthe two-hour practical round in your own IDE, judged on low-level design and on the code running; read through search snippets because Blind blocks direct reads
• 04TechInterview, what Glean's engineering interview actually testsstage durations, the two-hour build ('a lot of functioning code, not a pristine fraction of one'), the search design topics (hybrid indexing, permission-aware retrieval, freshness, ranking) and a three-to-five-week timeline; 20 July 2026, cites no candidate sources
• 05Glean careers, Software Engineer, Fullstackthe AI-focused exercise or discussion in every interview process, the hybrid office policy, and the $140K to $265K base range; read 23 September 2026
• 06Glean careers, Machine Learning Engineer, Search Qualityquery and document understanding, ranking signals, domain-adapted language models, four days a week in Mountain View, $140K to $265K base; read 23 September 2026
• 07Glean job board127 open roles on 23 September 2026, 26 of them engineering or data science, across Mountain View, San Francisco and Bangalore, with posted base ranges from $115K to $300K
• 08Levels.fyi, Glean software engineer, United Statesthe $222K US median and the $330K Bay Area median, read through search results on 23 September 2026
• 09TechCrunch, Glean lands a $7.2B valuationthe $150M Series F led by Wellington Management in June 2025, after a $260M Series E at $4.6B
• 10Wikipedia, Glean Technologiesthe founding in Palo Alto in 2019 by Arvind Jain, a former Google engineer

Interview processes change. This reflects widely-reported and sourced conditions as of 2026 — confirm specifics with your recruiter, and treat it as a map rather than a guarantee.