1. LandOffer
  2. Blog
  3. All articles
  4. GitLab Technical Interview Handoff: Finish the Async Merge Request Review Before the Live Call

Job search

GitLab Technical Interview Handoff: Finish the Async Merge Request Review Before the Live Call

A completed original review packet and timezone-aware handoff record distinguish MR preparation from the new generalist CodeSignal route.

9 min readLandOffer team

Confirm that your invitation uses the merge request route before preparing a GitLab MR review. GitLab's new generalist hiring page says its initial India-only process replaces the previous MR assessment with a CodeSignal take-home and two live technical interviews. A public MR guide is therefore not a universal description of every GitLab interview.

If your confirmed route does include the asynchronous MR, treat the written review as a handoff the interviewer needs before the call. Finish the reasoning, make the feedback available through the instructed workflow, and explicitly signal completion. Keep your broader search moving with current tech roles on LandOffer while checking the exact process for this application.

A candidate organizing a written technical review before a later call

First identify the route in your actual invitation

The generalist engineering page, added October 20, 2026, describes a process initially tested in India. It applies to roles categorized as generalist, with specialized roles using a different process. It describes CodeSignal assessment stages and team matching after a hire-and-level decision. That does not establish a worldwide replacement, nor does passing an assessment alone mean an offer exists.

For an invitation that names an MR, verify the exercise link, deadline, live-session format and permitted tools. For an invitation naming CodeSignal, use that route's supplied instructions rather than trying to locate a private MR. If the materials disagree, ask the recruiting contact which instructions govern your application. Do not resolve the disagreement by choosing whichever assignment seems easier.

This guide checked current public GitLab materials on October 21, 2026, including the technical guide last modified August 19 and the newly added generalist page. No candidate account, private exercise or live call was accessed. LandOffer publishes this guide and provides its own job-search product. All candidate records, code and messages below are original fictional examples, not GitLab assessment content.

Know what must be ready before the live call

The technical interview guide describes a self-contained exercise from a curated bank, rather than a request to contribute to a production issue. It says the MR link arrives at least 72 hours before the call, suggests spending up to an hour reviewing, and asks for completion at least 24 hours before the interview plus a platform note marking the asynchronous work finished.

The live component is described as a 90-minute video and screen-sharing session involving discussion of the review and coding improvements. Follow the actual invitation if its scheduling details need clarification. This article does not turn those public timings into a promise about when any individual candidate will receive materials.

Write a review the interviewer can understand before you are there to explain it. A long private notebook does not serve that purpose if the interviewer cannot see the findings. Equally, a short completion note is not a substitute for an actual review. The two artifacts do different jobs: one conveys your analysis, the other tells the reader that you have reached a stopping point.

Worked review: prioritize a reproducible correctness issue

Fictional candidate Marta practices on an original document-revision helper. The toy change claims to return the newest revision from records whose revision numbers are positive integer strings. It selects a maximum using the string field directly. With revisions "9" and "10", lexicographic comparison selects "9" rather than the numerically latest revision.

This is a local rehearsal fixture written for this article. It is not a real GitLab MR, predicted interview prompt or borrowed production code. A small Python check was actually executed locally: the flawed selection returned 9, and a numeric comparison returned 10. The result checks only those two records under the stated input contract.

Marta completes three review comments, kept as unsent examples here:

Priority Evidence from the fictional change Completed review comment
Correctness String maximum chooses revision 9 over 10 “The selection compares strings. With revisions 9 and 10, it returns 9. Could we compare the numeric revision and add this boundary case?”
Contract question Positive integer strings are assumed; invalid values are not specified “Should an invalid revision reject the input, be skipped, or be handled earlier? I would confirm that before adding silent coercion.”
Test coverage The example covers only the two-digit transition “Please retain a one-record case and check empty input and duplicate revision behavior against the agreed contract.”

Marta requests a targeted correctness fix and clarification of the invalid-input policy. She does not present every possible concern as a blocker. Nor does she propose replacing the storage layer to solve a comparison defect.

The comment contains a concrete input, observed output and intended behavior. “This looks wrong” would leave the author guessing. “Use integers” alone would omit the reason and could encourage a patch that throws unexpectedly on invalid input. Marta keeps the difference between the confirmed bug and the unanswered contract question visible.

She also records a rejected alternative: padding revision strings to a fixed width would only work if a maximum width were guaranteed. No such limit exists in this toy contract. Numeric ordering is a plausible minimal direction for valid records, but the helper still needs an agreed response to invalid and empty inputs. The local comparison check is not a complete application test suite.

Finish the review workflow, not just the text

GitLab's review documentation distinguishes unpublished review comments from submitted feedback. Starting a review or adding a comment to the review can leave it pending; submission publishes the review comments. The documented approval shortcut can also approve the MR, so do not use approval merely as a way to make comments visible when you have not reached that judgment.

Use the outcome and completion mechanism requested for the exercise. Read back the resulting review state before treating it as delivered. A browser draft, a local Markdown file and a submitted review are different states. This article did not publish comments, submit a review, approve a change or test any candidate project's permissions.

For Marta's rehearsal, the review packet is complete when the three comments, overall finding and known gaps agree. The material correctness issue remains explicit. Her completion statement would not say “approved” while her main finding still requires a change. The interviewer can then read her judgment without reconstructing it from scattered remarks.

A review packet and clock arranged to show an asynchronous handoff and later discussion

Worked completion record across a clock change

Consider Marta's hypothetical MR-route call at 09:00 America/Los_Angeles on Monday, November 2, 2026. That is 17:00 UTC. This example crosses the November daylight-saving transition; subtracting hours in UTC prevents a casual “three calendar days” calculation from shifting the intended lead time.

The following arithmetic was actually checked locally with Python's timezone-aware dates. These are invented schedule inputs, not a real invitation or proof of a platform submission.

Event Local time in the fictional record UTC and completed decision
Call Nov. 2, 09:00 PST 17:00 UTC; anchor for the calculation
Latest receipt consistent with 72 hours' lead Oct. 30, 10:00 PDT 17:00 UTC; distinguish this from the candidate's review deadline
Actual receipt in the example Oct. 30, 09:00 PDT 16:00 UTC; 73 hours before the call
Review cutoff at 24 hours before the call Nov. 1, 09:00 PST 17:00 UTC; deadline for the async work in this example
Marta's chosen finish Oct. 31, 16:00 PDT 23:00 UTC; 42 hours before the call

Marta's completed decision is to finish on Saturday afternoon, leaving 18 hours before the calculated review cutoff. She reserves the later time for reading her own findings and preparing to discuss them, rather than quietly expanding the assignment beyond the suggested review effort.

Her fictional completion note is fully drafted but unsent:

My asynchronous review is complete. I identified the revision-ordering issue with a two-record reproduction, left a question about invalid-input handling, and suggested contract-focused tests. The review does not establish empty-input or duplicate-revision behavior. I am ready to walk through the findings and work on the agreed improvement during the call.

The note identifies what is finished and what remains open. It does not claim the original change is production-ready or that a reviewer approved her reasoning. In a real exercise, she would check that the intended comments are available through the requested platform workflow before leaving this completion signal.

If the link arrives too late, access fails or the invitation's timezone is ambiguous, the sensible next step is clarification with the recruiting contact. Do not manufacture a completion timestamp or describe unpublished drafts as delivered. No such message or platform action was performed for this article.

Keep AI assistance attributable to the correct stage

GitLab's current candidate AI guidance welcomes transparent technical-assessment use while expecting candidates to explain prompts and decisions and avoid presenting generated content as their own. Its take-home guidance calls for citation and answers in the candidate's own words. Its interview guidance asks candidates to engage without reading AI-produced answers, and it expects disclosure of tools used.

That is a reason to preserve your reasoning and attribution, not to leave a hidden answer generator running during a live conversation. Clarify how those general guidelines apply to the supplied exercise, including which material may be shared with a tool. An assessment link does not automatically authorize uploading its contents to any external service.

Marta's fixture and review comments are original editorial examples, not captured output from a live coding assistant or candidate assessment. In an authorized rehearsal, a useful record would name any tool used, describe its limited task, and identify the conclusions the candidate personally verified. She should be able to explain why a suggested change was accepted or rejected without reading a generated explanation aloud.

A separate AI-native hiring working-group page labels its status and interview structure as proposed. Treating design proposals as proof of a deployed candidate process would confuse planning with implementation. Your invitation and the applicable candidate guidance still need to establish the route and stage requirements.

Prepare a concise live walkthrough from the same evidence

When the MR route is confirmed, practice describing the most consequential finding first. Marta starts with the mismatch between the numeric contract and string comparison, gives the two-record example, and explains the smaller repair direction. She then separates the invalid-input question from the reproduced bug.

If the interviewer changes the contract, she can revise the recommendation instead of defending the old answer mechanically. For example, if revisions are not numeric identifiers at all, her original ordering assumption needs to be reconsidered. That is a discussion about requirements, not a failure to memorize the right comparator.

Bring a clear account of what you examined, what you ran and what you did not check. Leave space for collaboration: an interviewer may want to explore a different improvement rather than execute your entire wish list. The async packet should help that conversation begin with shared context.

Confirm the route, deliver the written review and completion signal, and prepare to discuss the same evidence live. Keep that preparation connected to your actual experience and relevant roles in your ongoing search, rather than assuming every GitLab application follows the same MR handoff.

Sources and Further Reading