
Two candidates describe different American Express ML Engineer interviews. One centered on GenAI project code and LLM fundamentals such as transformers and fine-tuning. The other was a thinner conversation in which the candidate asked about the team's work and PII handling.
$124K
Avg. Base Comp
$140K
Avg. Total Comp
Not reported
Typical Rounds
Not reported
Process Length
The two candidate reports for the American Express ML Engineer interview differ, but both point toward generative AI work. One candidate had a virtual first round built around a personal GenAI project. The interviewer had read the code and, in place of a leetcode-style problem, asked the candidate to write code based on that project. If you list a project on your resume, be ready to explain its design choices and extend it live.
That same candidate described the on-site as roughly an hour and a half of questions on LLM foundations: how LLMs work, transformer architecture and fine-tuning. One question asked how to cope with running out of a paid LLM usage limit. Another asked the candidate to talk through the transformer. The candidate was not offered the role and was told the company chose internal candidates already familiar with RASA, the framework used for its chatbots. Familiarity with conversational AI tooling may therefore help.
The second candidate reported a thinner technical conversation and spent much of it asking the interviewers questions. They asked about the team's technology and models, and about how PII is protected when sensitive data goes to the cloud. They felt the answers blurred pre-upload masking with output guardrails. That candidate declined an offer. If data handling matters to you, separate masking before upload from guardrails on model output, and ask the team for concrete answers. Experiences vary, so prepare for a hands-on project discussion and also come ready with your own specific questions about the team's work.
Synthesized from 2 candidate reports by our editorial team.
Had an interview recently?
Share your experience. Unlock the full guide.
Real interview reports from people who went through the American Express process.
I didn't get the role, and the feedback was odd. They pointed to no real weaknesses in my interviews, but they went with internal candidates who already knew RASA, the framework they use for their chatbots. Going in, I didn't expect my own side projects to matter so much.
The first round was virtual and mostly me walking through a personal GenAI project. The interviewer had clearly read my code. When it came time for the leetcode-style problem I was supposed to get at the end, he asked me to write code based on the project instead.
The on-site was the hard part. I was grilled for about an hour and a half straight, mostly on the foundations of LLMs: how they work, their architecture, and fine-tuning. Two questions stuck with me. One asked how I would mitigate running out of a paid LLM usage limit. The other asked me to talk through the transformer architecture.
If you're preparing for something like this, treat your project as an interview in itself, since they will read your code. Be ready to explain the design choices behind it, and go deep on transformer internals and fine-tuning rather than skimming them.
Prep tip from this candidate
Be ready to walk through your own project's code live and explain your design choices, since the interviewer may ask you to write code based on it. Also go deep on transformer architecture and fine-tuning, and have a plan for managing cost and usage limits on a paid LLM API.
Share your own interview experience to unlock all reports, or subscribe for full access.
Sourced from candidate reports and verified by our team.
Topics based on recent interview experiences.
| Question | |
|---|---|
| P-value to a Layman | |
| Assumptions of Linear Regression | |
| Bagging vs Boosting | |
| Regularization and Validation | |
| Random Forest from Scratch | |
| Find the First Non-Repeating Character in a String | |
| Singly Linked List | |
| Z and t-Tests | |
| Decision Tree Evaluation | |
| Overfit Avoidance | |
| Precision and Recall | |
| Hurdles In Data Projects | |
| Xgboost vs Random Forest | |
| Random Forest Expansion | |
| Variable Error | |
| Loan Model | |
| 85% vs 82% | |
| Text Editor With OOP | |
| User Event Data Pipeline | |
| Fixed-Length Arrays: Deletion | |
| Acquisition Threshold | |
| Why Do You Want to Work With Us | |
| Client Solution Pushback | |
| Your Strengths and Weaknesses | |
| Open Source Reporting Pipeline | |
| Stakeholder Communication | |
| Optimize Model Performance | |
| Different Card | |
| Subscription Overlap |
Synthesized from candidate reports. Individual experiences may vary.
One candidate reports a virtual first round spent mostly walking through a personal GenAI project. The interviewer had read the code and, instead of a leetcode-style problem, asked the candidate to write code based on the project. Prepare to defend design choices and extend your own code live.
The same candidate describes an on-site of about an hour and a half focused on LLM foundations, including transformer architecture and fine-tuning. One question asked how to mitigate running out of a paid LLM usage limit, so cost and usage planning may come up for GenAI roles.
A different candidate reports a thinner technical conversation in which they questioned interviewers about the team's technology, its models and how PII is protected when sensitive data is uploaded to the cloud. They felt answers on masking before upload were unclear. Candidates may want to ask concrete questions about the team's work and data handling.