Gimlet Labs · Phone screen → Onsite
In processGimlet Labs Experienced MLE Phone Screen and Onsite: 1 Experience
A Gimlet Labs experienced MLE loop in late 2026: an LRU cache screen, then a dashboard build, FLOPs-and-bytes modeling and an inference serving design.
At a glance
- Process: phone screen (coding) → three-round onsite (design then build, performance modeling, system design) → a casual founder chat (report 01).
- Asked recently: LRU cache; a utilization dashboard from two unaligned metric series, built from scratch; counting FLOPs and bytes; an inference serving system (report 01, report date Nov 2026).
- Difficulty: rated easy; the dashboard round had no framework or tests. Passed.
- Practise first: LRU Cache and Design a Large-Model Inference Serving Platform.
Coverage: One experienced machine learning engineer with a Bachelor's degree, changing jobs, who came to Gimlet Labs through a recruiter. The report covers a phone screen and a three-round onsite plus a chat with a founder. It gives October to December 2026 as the interview period and went up on November 12, 2026.
What came up
The loop mixed a classic coding question with work close to running AI inference: an open-ended dashboard to build from scratch, a performance-modeling round on FLOPs and bytes, and an inference serving design.
| Topic | Reports | Last seen |
|---|---|---|
| Coding: LRU cache (phone screen) | 1 of 1 | Nov 2026 (report date) |
| Design then build: a utilization dashboard from two unaligned metric series | 1 of 1 | Nov 2026 (report date) |
| Performance modeling: counting FLOPs and bytes | 1 of 1 | Nov 2026 (report date) |
| System design: an inference serving system | 1 of 1 | Nov 2026 (report date) |
Questions we found
Algorithms / Data Structures
| Question | Reports | Last seen |
|---|---|---|
| Implement an LRU cache (LRU Cache) | 1 | Nov 2026 (report date) |
Coding and Design
| Question | Reports | Last seen |
|---|---|---|
| Given two sets of time-based count metrics that aren't aligned, design and then build a dashboard showing how utilization changes over time, with no starter framework or tests | 1 | Nov 2026 (report date) |
System Design
| Question | Reports | Last seen |
|---|---|---|
| Design an inference serving system (Design a Large-Model Inference Serving Platform) | 1 | Nov 2026 (report date) |
Performance Modeling
| Question | Reports | Last seen |
|---|---|---|
| Routine performance modeling: count FLOPs and bytes | 1 | Nov 2026 (report date) |
1 detailed candidate timeline
01: MLE phone screen and a three-round onsite
Interview period: 2026 Q4 · Report date: November 12, 2026 · Role: Machine learning engineer · Degree listed: Bachelor's · Channel: Recruiter · Candidate's rating: Positive, easy · Result: Passed, per the report
The candidate describes Gimlet Labs as an inference cloud that has been raising money fast, builds its own data centers and runs several kinds of AI chips.
Phone screen
- Coding: an LRU cache. The report names the exercise and nothing more. FastPrep's Gimlet Labs version is LRU Cache; its operation format and limits are FastPrep's practice choices.
Onsite round 1: design, then code
- The candidate was given two sets of time-based count metrics that don't line up in time and asked how they would design a dashboard showing utilization change over time.
- After the design discussion, they wrote the solution. There was no framework and there were no test cases; the candidate had to come up with everything and get it running. The report doesn't give the data format or the exact output, so there is no practice version of this round.
Onsite round 2: performance modeling
- A routine performance-modeling round. The candidate says being able to count FLOPs and bytes was enough.
Onsite round 3: system design
- Design an inference serving system. The prompt was very open, and the candidate chose which directions to extend it in. FastPrep's practice version, which lists Gimlet Labs among its companies, is Design a Large-Model Inference Serving Platform. The report gives no requirements, so the practice requirements and scale are FastPrep's.
Founder chat
- A casual conversation with a founder.
How it went
The candidate enjoyed the process. Every interviewer was friendly, and the founder came across as down to earth.
Common questions
What is the Gimlet Labs MLE onsite?
Three rounds in this report: design and then build a utilization dashboard from scratch, a performance-modeling round on FLOPs and bytes, and an inference serving system design, followed by a founder chat (report 01).
What coding question does Gimlet Labs ask in the phone screen?
An LRU cache (report 01).
Keep practicing
- Coding: LRU Cache is the practice version of the phone screen. More Gimlet Labs questions, in every practice format, are in the Gimlet Labs question bank.
- System design: Design a Large-Model Inference Serving Platform is the practice version of onsite round 3.
- More firsthand reports: the Experience feed.
Sources: Anywhere useful interview info can be found.
Comments
0No comments yet — be the first to share what you know.