Adobe ML Engineer Round 1: 3 Experiences (Part 11 of 12)
- Company
- Adobe
- Reports
- 3 interview reports
- Level
- Full-time
- Stages
- Phone screen
Adobe machine learning engineer round 1 reports from 2025–2026 lead with ML system design; two added coding, a training loop or a short DSA problem.
At a glance
- Process: one technical round of about an hour (66). Candidate 65 had both technical interviews on consecutive days. Later rounds were hands-on ML on a dataset (65), a harder scenario problem (66), or Python coding, an LLM deep dive and managerial rounds (67).
- Asked: an end-to-end ML design in 2 of 3 (65 fraud detection, 67 a document's aspect ratio), a project deep dive in 1 (66), a neural-network training loop in 1 (65), and an easy-to-medium DSA problem in 1 (66).
- Difficulty: candidate 66 solved the DSA problem in about 10 minutes. One of three got an offer (67).
- Practise first: Flag Fraudulent Subscription Accounts From Event Logs and Predict a Document's Aspect Ratio From Its Images and Text.
Coverage: 3 reports from 2025 to Nov 2026: Machine Learning Engineer 3 on the Genuine team (65), an AI/ML engineer in India (66) and Machine Learning Engineer II (67). The series covers 69 experienced-hire reports, first technical round only.
In this series
- Part 1: overview — the process, most-asked problems and a guide to every role
- Part 2: MTS-2, Jun 2025–Sep 2026 — 8 MTS-2 reports, entries 01–08
- Part 3: MTS-1 and MTS-2, Nov 2024–May 2025 — 7 more MTS reports, entries 09–15
- Part 4: Computer Scientist 1, Jul 2025–Aug 2026 — 8 CS-1 reports, entries 16–23
- Part 5: Computer Scientist 1, Dec 2024–Aug 2025 — 7 more CS-1 reports, entries 24–30
- Part 6: CS-2 and Senior Computer Scientist — 4 reports, entries 31–34
- Part 7: SDE, SE2 and senior engineers — 8 reports, entries 35–42
- Part 8: C++ engineers — 9 reports, entries 43–51
- Part 9: frontend engineers — 10 reports, entries 52–61
- Part 10: Android engineers — 3 reports, entries 62–64
- Part 11 (current): machine learning engineers
- Part 12: release engineering and product consultant — 2 reports, entries 68–69
What came up
Two of the three first rounds were an ML design discussion rather than coding.
| Topic | Reports | Last seen |
|---|---|---|
| ML system design | 2 of 3 | Nov 2026 (report date) |
| Neural-network training code | 1 of 3 | Nov 2026 (report date) |
| Project deep dive | 1 of 3 | Aug 2025 |
| DSA problem | 1 of 3 | Aug 2025 |
| Computer vision and AI questions | 1 of 3 | Aug 2025 |
Questions we found
Machine Learning
| Question | Reports | Last seen |
|---|---|---|
| Flag Fraudulent Subscription Accounts From Event Logs: design a fraud-detection ML system | 1: 65 | Nov 2026 (report date) |
| Define a network and write its training loop for multi-label prediction (PyTorch or TensorFlow) | 1: 65 | Nov 2026 (report date) |
| Predict a Document's Aspect Ratio From Its Images and Text | 1: 67 | Oct 2025 (report date) |
3 ML engineer round 1 reports
65: MLE 3 on the Genuine team, fraud-detection design plus PyTorch
Report date: Nov 7, 2026 · Role: Machine Learning Engineer 3 (Adobe Genuine team) · Experience: 2.5 years, plus a PhD in ML · Result: No reply after follow-ups; the application still showed as under review
Round 1: ML system design and PyTorch coding (first technical round)
An outside recruiter reached out, and both technical interviews were set for the next day.
- Design a fraud-detection ML system. The talk covered data collection, feature engineering, outlier detection, class imbalance, model choice, deployment, and online vs offline metrics. No scale, input schema or required algorithm was given.
- Define a neural network and write its training loop for multi-label prediction. The candidate could pick PyTorch or TensorFlow. No dataset was given, so the code wasn't run. The exact labels, loss and layer sizes weren't shared.
After round 1: Round 2 was hands-on ML on an event-level CSV with about 15 features and a fraud label. The interviewer nudged the candidate to build user or account level aggregates. The candidate wanted to drop the account ID because of encoding worries and got stuck. The last 15 minutes covered model choice and metrics.
66: AI/ML engineer, projects first, then a quick coding problem
Interview date: about Aug 2025 (about three months before the Nov 16, 2025 report) · Role: AI/ML Engineer · Location: India · Result: Cleared round 1; rejected after round 2
Before the first technical round: An HR call asked about total experience, current role, notice period and expected pay.
First technical round (about 1 hour)
- A long, detailed talk about the candidate's projects and what they built.
- An easy-to-medium DSA problem. The candidate solved it in about 10 minutes while explaining the logic. The problem itself wasn't named.
- Computer vision and AI questions; the exact questions weren't shared.
HR confirmed the same day that the candidate had cleared the round.
After round 1: The second, more advanced round had a scenario-based problem where the candidate needed hints, plus one simple DSA question. They didn't clear it and were told about a six-month wait before reapplying.
67: MLE II, predicting a document's aspect ratio
Interview date: 2025 (report date Oct 12, 2025) · Role: Machine Learning Engineer II · Experience: 3+ years as a data scientist and MLE · Result: Offer, accepted
Round 1 (first technical round)
- End-to-end ML design for documents that have both image and text data: propose a full pipeline that predicts a document's aspect ratio.
This was the only question reported for the round.
After round 1: Round 2 had one Python coding question and a deep dive on LLMs, transformers and attention. Round 3 was managerial (projects, teamwork, behavioral). The hiring manager round asked for a design of a color suggestion system. The candidate was selected and accepted.
Common questions
Does the Adobe ML engineer interview include coding?
In round one, two of three did: a training loop in PyTorch or TensorFlow (65) and a short DSA problem (66). Later rounds added hands-on modeling on a CSV (65) and a Python coding question (67).
Keep practicing
- Practice from these reports: Design a fraud-detection ML system (65); predicts a document's aspect ratio (67).
- Adobe coding bank: every Adobe algorithm problem in the question bank (sign-in needed).
- Adobe ML modeling bank: the ML modeling questions.
- Adobe overview: the Adobe interview page and Adobe hiring insights.
- More reports: the Experience feed.
In this series
Sources: Anywhere useful interview info can be found.
Comments
0No comments yet — be the first to share what you know.