Harvey ML Operations Engineer Interview
Updated by Harvey AI candidates
Written by Graham Carlson, Senior Technical ContributorOur guides are created from recent, real, first-hand insights shared by interviewers and candidates. If your experience differs, tell us here.
Harvey is a startup that creates AI tools to help legal professionals streamline every aspect of their work.
Their ML Operations Engineers are dedicated to building sophisticated infrastructure to support Harvey’s experiments, product design, and platforms.
Their interview rounds are practical, hands-on assessments and discussions. You’re given a lot of freedom to use any tool and encouraged to draw from your past work to show off your skills.
There are no explicit cultural or behavioral rounds; the technical questions are designed to showcase your ability to collaborate and communicate cross-functionally.
Below, we break down the complete ML operations engineer interview process at Harvey.
Here’s a 1st-hand account from a Harvey Interview: “There are several unique aspects about this interview process, including:
- Focus on experimentation and ideation, searching for novel ways to solve persistent problems
- Questions requiring deep architectural and infra knowledge, not just ML concepts
- Testing to see how you handle the pressure.”
Interview Process
Harvey’s interview process will focus primarily on technical skills and solutions architecture, but you will also be given chances to discuss the organization’s goals, structure, and approach to work.
Each round will be remote, and the rounds may all be scheduled on different days. Every round after the recruiter screening call is 60 minutes, but some rounds may go longer.
- Recruiter screen: A short call to discuss the role, your past projects, and experience with ML operations
- Coding assessment: A one-hour take-home assignment to assess your coding skills
- Paired coding interview: An AI-focused coding assessment that will pair you with a Harvey engineer
- Solution architecture presentation: A discussion with a panel of engineers about a past project or a proposed solution to a hypothetical problem
- Director interview: A high-level discussion about the role and the future direction of the organization
Recruiter screen
This brief interview will focus on your experience creating performant, stable systems and production environments. They’ll also touch on the role and may gauge your expectations for salary and benefits.
As the creator of popular AI tools, Harvey is becoming increasingly focused on system reliability, and they’ll want to know your experience testing and building scalable tools.
Recruiter Questions:
Asked at 
Asked at 
Asked at Coding assessment
This will be a timed, take-home-style assessment that you will have an hour to complete sometime after the recruiter screen.
Unlike the other rounds, this assessment will not necessarily have much to do with Harvey’s products or infrastructure, but rather be a test of your general problem-solving skills.
Some standard engineering practice prep can help you master these generalized engineering questions.
Also, unlike the other rounds, you won’t have much tool flexibility and will be required to complete the assessment in one go without leaving the tab.
Coding assessment questions:
Asked at Adobe, Apple, Goldman Sachs + 6 more Asked at Adobe, Apple, Capital One + 11 more Paired coding interview, 60 minutes
In this round, you’ll be paired with a technical team member, and you’ll work through a coding problem and follow-up questions using screenshare.
The problem will be practical and related to ML principles, such as tokenizing a string and storing it in a hypothetical vector database.
It may start with what appears to be a standard data structures and algorithms problem, such as a tree or graph problem, but with a greater focus on systems and techniques like data embeddings.
The interviewers at Harvey tend to be stingy with hints, which follows their approach of using each question as a mini-behavioral test. In this case, they’re assessing your ability to think through a problem autonomously and function under pressure.
You’ll be asked lots of open-ended questions, and you’ll be given the freedom to work out a solution using any tools you prefer, including LLM coding assistants. They will provide a validator tool for you to test your solution.
Paired Coding Interview Questions:
Asked at Harvey AI • 
Asked at Databricks 
Asked at Snap • Solution architecture presentation, 60 minutes
This round is a panel discussion with multiple technical team members, typically ML engineers and AI researchers. The time limit is 60 minutes, but the discussion can sometimes go longer at the panel’s discretion.
You’ll be asked to either demonstrate the architecture of a past project or propose architecture for a hypothetical new project. It’s best to choose something similar to the products and services they offer or that address a problem common to LLM-based products.
Like the prior round, you’ll be able to share your screen and present your solution in any format, such as a slide deck or interactive chart.
As you go through your presentation, they will ask questions about your choices, what your goals were, which models you built and shipped, and why you chose one solution over another. They will want to see your awareness of the trade-offs and that you made well-informed choices.
They will also ask hypotheticals, such as how you would wring more speed or performance out of a particular component, how you would handle requests from different AI researchers, engineers, and other stakeholders, and how you would build a stable, scalable system.
Solution Architecture Questions:
Asked at Director interview, 60 minutes
This interview is one-on-one with an executive at Harvey. It will be an informal discussion about the role and larger questions about the company and its goals.
You’ll be prompted to ask lots of questions of your own to explore whether this role is a good fit. They’ll want to gauge your excitement about the potential of their tools and your approach to work more generally.
Although it has grown into a medium-sized organization, Harvey is known for encouraging all team members to experiment and propose new solutions. They will ask about your experiences in this kind of environment and how you choose which experiments to build up into features.
It’s important to express your excitement about Harvey’s approach and to clarify that you understand the role, which is primarily geared toward supporting their ML engineers and researchers as they develop new products and features.
Director Interview Questions:
Asked at 
Asked at Common Mistakes
Here are some of the mistakes people make when preparing to interview for an ML operations engineer position at Harvey:
- Over-preparing for culture and behavioral questions, as these will not be addressed directly by Harvey’s interviewers
- Choosing a past project that isn’t fresh in your memory or relevant for the solutions architecture round, as you will be expected to answer a large number of follow-up and hypothetical questions
- Expecting lots of hints and information before and during each round, as part of Harvey’s behavioral assessment, is to see how you function with limited input
- Assuming that Harvey, as a startup, is interested in hacky or brute force approaches. Because of their size and popularity, they are at a stage which is more interested in stability and scalability
- Not preparing for questions that explore the details of infrastructure work, as database performance optimization is a common follow-up question
- Failing to show excitement or interest in Harvey’s products and approach, as this is the primary focus of the director interview
Interview Prep
Demonstrate balance: As a hybrid position of operations and ML engineering, you’ll be expected to be able to dig into the details of data infrastructure and to build complex architecture to support Harvey’s feature development. You’ll need to demonstrate your ability to improve infrastructure performance and adapt to different use cases for researchers and engineers.
Adjust your work style: Because they do not have a defined behavioral assessment, you’ll need to show your collaborative and cross-functional skills through your work. In addition to your practice prep, think through the actual use cases of what you are building, how you would explain them to non-technical people, and how they would evolve as Harvey’s goals change.
Understand Harvey’s position: Harvey is a maturing startup working in the legal world, which means that stability and reliability are very important to them. You’ll need to demonstrate that you understand the details of provisioning infrastructure for scale and performance, not just speed.
About the Role
Harvey’s ML Operations Engineers are in a unique position to support critical infrastructure while also driving ongoing research and development of new features. Here are some common characteristics:
Build infrastructure to support new development: You’ll work with their research and ML teams to develop performant data storage, processing, and monitoring to ensure they have a stable and reliable development environment.
Continually assess and improve infra performance: As an operations specialist, you’ll use performance monitoring and analysis to identify potential issues and opportunities for more performant, stable, and consistent infrastructure.
Experiment to develop new approaches: Your work with researchers will leverage their findings to improve existing infrastructure and test novel solutions to existing or emerging issues. You’ll also be empowered to experiment and propose new ideas to help with your work or other teams at Harvey.
Core Responsibilities
As an operations expert at a maturing startup, your job will focus on building the data and processing tools that support product development. Here are some of the general tasks you’ll be asked to complete for this role:
- Create, monitor, and improve Harvey’s data and model infrastructure to support the needs of the AI researchers, ML engineers, and other teams
- Create new infrastructure in accordance with the specific requirements of different projects, and then find ways to incorporate this tooling to Harvey’s solutions
- Continually find new opportunities to improve infrastructure efficiency and performance, particularly for model training
- Find ways to make the feature development experience more stable without limiting the flexibility of researchers and engineers to use their preferred tools and versions
Compensation
Harvey’s ML Operations Engineers make an average of $405,000/yr in total compensation. We’ve mainly seen packages in between the 300s and 500s.
Job Requirements
Experience
Applicants for this role are expected to have 10+ years of experience with machine learning infrastructure and platforms. Although it isn’t required, it’s very helpful to have experience working with AI/ML tools and inference.
Education
Although they do not list specific educational requirements or certifications, many companies in the AI space prefer to hire candidates with advanced degrees in relevant fields, such as computer science or data science.
Resources
- Harvey Careers
- Harvey Company Page
- Harvey Interview Questions
- Machine Learning Interview Prep
- Generative AI Interview Prep
- Machine Learning Interview Questions
FAQs
How long is the Harvey ML Operations Engineer interview process?
The exact timeline will depend on the interviewer's availability, but candidates report the process takes 4 to 6 weeks.
Does Harvey have internships?
They offer 12-week internships, primarily for technical roles. They post these on their careers page.
Does Harvey offer remote work opportunities?
Harvey operates with a specific hybrid office requirement of three days in the office. However, they do occasionally allow for fully remote work, depending on the role. They are hiring globally, but their main office is in San Francisco.
Do I need AI experience to work at Harvey?
It isn’t required, but the hands-on nature of the interviews and the heavy focus on AI mean you need to be familiar with the specific requirements of AI data infrastructure.
Do I need legal experience to work at Harvey?
No legal experience is required to get a technical role at Harvey.
How long should I wait after a rejection before reapplying to Harvey?
They do not provide a specific timeline on their site, but the best practice is to wait at least 6 months to a year before reapplying. If you were rejected for a specific reason, such as relocation issues, you can reapply sooner if that reason changes.
Learn everything you need to ace your ML Operations Engineer interviews.
Exponent is the fastest-growing tech interview prep platform. Get free interview guides, insider tips, and courses.
Create your free account