
What to expect in a Tower Research Capital ML Engineer interview, based on one candidate's report: a HackerRank screen mixing ML theory and coding, then ML system design, hands-on ML and a final leadership conversation.
$200K
Avg. Base Comp
$300K
Avg. Total Comp
3-4 rounds
Typical Rounds
Not reported
Process Length
One candidate's report, from a machine learning internship interview, describes a process that tests applied ML skill as much as coding. It opened with a HackerRank assessment of twelve multiple-choice questions on CS theory and machine learning, followed by a LeetCode-medium problem, a Python task and a bash script, all under time pressure.
The main loop that candidate describes had three to four rounds, ending with a conversation with a global head. One round was a system design case study grounded in a real use case. The candidate had to justify which LLMs to integrate and which vector database to use. Another round pressed on hands-on experience: CNNs were discussed in terms of building and deploying models at scale in production, not just theory. Classical ML, NLP, LLMs, Docker, Git and bash also came up.
The practical takeaway is to prepare concrete, shipped projects you can defend decision by decision. Practice turning a vague problem into a design with clear trade-offs, and be ready to show comfort with everyday production tooling. The final leadership round was described as more conversational, focused on impact and how you approach hard problems. This is a single internship report, so senior ML Engineer loops may differ.
Synthesized from 1 candidate report by our editorial team.
Had an interview recently?
Share your experience. Unlock the full guide.
Real interview reports from people who went through the Tower Research Capital process.
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.
Featured question at Tower Research Capital
Given two sorted lists, write a function to merge them into one sorted list.
| Question | |
|---|---|
| Subscription Overlap | |
| Find the Missing Number | |
| P-value to a Layman | |
| Get Top N Frequent Words | |
| Assumptions of Linear Regression | |
| Maximum Profit | |
| Duplicate Rows | |
| Bagging vs Boosting | |
| String Shift | |
| Scrambled Tickets | |
| LRU Cache 1 | |
| Lasso vs Ridge | |
| Weighted Average Sales | |
| Hurdles In Data Projects | |
| Priority Queue Using Linked List | |
| HHT or HTT | |
| Bias vs. Variance Tradeoff | |
| Flipping 576 Times | |
| 500 Cards | |
| Regularization and Validation | |
| Random Forest from Scratch | |
| Shortest Path Algorithms | |
| Find the First Non-Repeating Character in a String | |
| Type-ahead Search | |
| Level Of Rain Water In 2D Terrain | |
| Median Probability | |
| Overfit Avoidance | |
| The Brackets Problem | |
| Singly Linked List |
Synthesized from candidate reports. Individual experiences may vary.
The candidate reports a HackerRank assessment with twelve multiple-choice questions on CS theory and machine learning. It also had a LeetCode-medium problem, a Python coding task and a bash script, all timed. It acted as a filter for coding ability combined with ML knowledge.
One interview round gave this candidate a system design case study. They had to architect a solution and justify choices such as which LLMs to integrate and which vector database to use. It was reported as grounded in real team problems rather than a generic design prompt.
Another round, per the same candidate, drilled into applied experience. CNNs came up, with the expectation of discussing building and deploying models at scale in production. Classical ML, NLP, LLMs, Docker, Git and bash were also covered, so be ready to talk through real projects.
The candidate describes the loop as three to four rounds in total, ending with a global head. This last conversation felt more like a discussion of impact and how you approach hard problems than a technical grilling. The candidate reports receiving an offer.