Meta's machine learning engineer (MLE) interview changes with the role's specialization: generalist MLE candidates and specialized candidates, like computer vision engineers, get different coding and design rounds. At Meta, "AI coding" can mean an LLM-assisted debugging session in an unfamiliar codebase or a set of ML math implementations like sparse matrix-vector multiplication, depending on your track. Every version of the loop tests how quickly you write correct code under time pressure, which makes timed coding practice the foundation of your prep.
This guide breaks down each stage of the Meta MLE interview process, what interviewers look for, and how to prepare with real example questions, actionable tips, and resources.
Meta MLE interview process
The Meta MLE interview process includes a recruiter screen or written pre-screen, a technical screen, a final loop, and team matching. Meta posts generalist MLE roles under the title "Software Engineer, Machine Learning," and specialized tracks follow a modified version of the process.
- Recruiter screen or pre-screen: A written questionnaire on Meta’s candidate portal, a recruiter call, or a hiring manager screen on specialized tracks
- Technical screen: A 45-minute live coding interview with 2 challenges, or a 60-minute AI-native screen for some senior candidates
- Final loop: 4-5 virtual interviews covering coding, design, and behavioral, sometimes split across several days
- Team matching: Placement on a specific team after you pass the final loop
Expect roughly 4-8 weeks from first contact to a hiring decision, and plan for team matching to add more time.
Meta now builds an AI assistant into many of its interviews, including coding and design rounds. Use the CoderPad practice environment in your Meta Career Profile to get comfortable with it before your screen.
Here's how the final loop compares across generalist and specialized tracks, such as a recent computer vision engineer loop:
Recruiter screen and pre-screen questionnaire
Meta's MLE pre-screen confirms that your experience matches the target level before any technical interview. Depending on the loop, it’s a written questionnaire on Meta’s candidate portal, a recruiter call, or a screen with a hiring manager.
The portal questionnaire is mostly open-ended written questions about your current team, tools, and role, with a few multiple-choice items. Answer in full paragraphs with specific details on team size, scope, and tools, since Meta uses these answers to decide whether to move you to the technical screen.
For many E5 and above candidates, the questionnaire centers on how you use AI tools day to day, how you keep up with new tooling, and how you judge AI output quality. Answer with specific tools, workflows, and examples of AI mistakes you caught.
Recruiter calls verify your years of relevant experience against the level you’re targeting. Have specific examples ready that match the scope expected at that level.
On specialized tracks, the hiring manager asks about each past project, what you delivered, and its impact. Prepare a summary of each recent project that ends with a measurable result.
Interviewers look for:
- Level fit: Whether your years of experience and scope match the target level
- Project impact: What you delivered on past projects and how you measured the result
- Technical depth: How well you explain the motivation, assumptions, trade-offs, and limitations behind your past work
- Tools and AI use: Which tools, programming languages, and AI tools you use in your current role
- Role mix: How you divide your time between hands-on coding and leadership
Recently asked questions
Here are some real interview questions reported by candidates:
- What's the size of your current team, and what does it work on?
- Which tools, programming languages, and AI tools do you use day to day?
- What percentage of your current role is coding vs. leadership?
- Walk me through each of your recent projects. What did you deliver, and what was the impact?
- How many years of experience do you have that qualify you for this level?
- How do you use AI tools day to day, and how do you judge the quality of their output?
Technical screen
The Meta MLE technical screen is a 45-minute live coding interview in CoderPad where you solve 2 data structures and algorithms challenges while explaining your reasoning. Expect to need complete, bug-free solutions to both challenges to advance.
Code execution is off in the standard screen. Trace the first solution line by line before you start the second challenge.
Save time at the end to test both solutions against edge cases and state their runtime and space complexity. Senior candidates may also get behavioral questions in the screen, so have 2 stories ready.
Some E5 and above candidates take a 60-minute AI-native screen with an AI assistant in place of the standard format. Ask your recruiter which version you’re scheduled for.
Interviewers look for:
- Presentation clarity: How clearly you explain your approach as you work
- Code traceability: Whether you can walk through your code and show exactly how it works
- Complexity analysis: Whether you can state the runtime and space complexity of your solution
- Code quality: How cleanly you write and present your code
- Correctness: Whether both solutions run without bugs
Recently asked questions
Here are some real interview questions reported by candidates:
- Given a Unix-style file path containing slashes and dots, determine whether it's a valid path.
- Given 2 sorted lists of intervals, merge them into a single sorted list.
- Given strings A and B, return every character in A that also appears in B.
- Find the shortest path between 2 points in a grid.
- Merge 2 sorted linked lists into a single sorted linked list.
- Find the k closest points to the origin.
- Count how many times a target integer appears in a sorted array in O(log N) time.
Coding interview
The Meta MLE coding interview is a 45-minute data structures and algorithms round in the final loop, where you typically solve 2 challenges from Meta’s question bank in about 40 minutes. Interviewers may state the target question count at the start and push you to move quickly between challenges.
If time is short, expect to explain your approach to the final challenge without coding it. Choose your programming language at the start of the round, and get each solution working before you optimize it.
Expect medium to medium-hard challenges, including modified versions of well-known questions. Meta doesn’t ask dynamic programming questions in these rounds, so put that prep time into trees, graphs, heaps, and string parsing. Read each prompt for what differs from the standard version before you apply a solution you’ve practiced.
Interviewers may add a follow-up constraint once your first solution works. Leave time to adapt each solution before you move to the next challenge.
Prioritize the questions most frequently reported for Meta in your prep.
Interviewers look for:
- Problem solving: How you break down each challenge and optimize for efficiency
- Coding: How clearly and correctly you implement each solution
- Verification: How you test and debug your own code, including catching bugs before the interviewer does
- Pace: Whether you complete multiple challenges inside the time limit
- Communication: How well you explain trade-offs, assumptions, and complexity
Recently asked questions
Here are some real interview questions reported by candidates:
- Count the mismatched brackets in a string that contains round, square, and curly brackets.
- Given a tree and 2 nodes, find their first shared ancestor. Can you find it without traversing to the root?
- Track the k largest elements of a running list.
- Group the nodes of a binary tree by vertical column.
- Return the index of the largest number in a list, choosing one index at random when the maximum appears more than once.
- Given a string of directory commands that uses periods, double periods, and tildes, determine the current directory.
- Given a sorted array and an element k, find the next element greater than k.
- Decide whether a string can become a palindrome after deleting at most one character.
Meta AI-enabled coding interview
Meta’s AI-enabled coding interview, also called the AI-assisted coding round, is a 60-minute MLE round where you implement and debug code in a multi-file CoderPad project with an AI assistant built into the editor. Most generalist MLE loops include it in place of one of the standard coding rounds, though some still run 2 standard rounds and others run 2 AI-enabled rounds.
The interviewer opens by explaining the challenge and the CoderPad layout, including a model picker for choosing among several LLMs. Expect several short files, test case files, and one core function left for you to implement, along with helper functions you may not need.
The task typically moves through 3 phases: fixing a bug, implementing the core logic, and optimizing against larger test data. Prioritize a correct core implementation, since finishing the final phase isn’t always required to pass.
The round supports 9 languages, including Python, Java, C++, and TypeScript. Confirm your language with your recruiter and practice in the same CoderPad environment before the loop.
Meta expects you to use the assistant, but it doesn’t score prompting skill on its own. Use it for well-defined subtasks like unfamiliar syntax or a known data structure.
The assistant can be less capable than the tools you use at work. Plan the algorithm yourself and treat its output as a draft.
Interviewers look for:
- Codebase navigation: How quickly you orient yourself in unfamiliar code and isolate failing tests
- AI judgment: Whether you verify the assistant's output and catch hallucinated functions or parameters
- Code reuse: How well you extend the existing code without rewriting working parts
- Complexity awareness: Whether you can state the time and space complexity of your solution and identify where it can improve
- Reasoning: How clearly you explain why a test fails and how your fix resolves it
Recently asked questions
Here are some real interview questions reported by candidates:
- Implement the core function in a multi-file codebase so every provided test file passes.
- Given a list of words, return the words that contain another word in the list as a substring.
- Your solution passes on alphabetic input. How would you extend it to handle numeric and mixed alphanumeric input?
- What's the time complexity of your solution, and can you make it more efficient?
ML coding interview (AI coding)
Meta's ML coding interview, which recruiters may label "AI coding," is a specialized-track MLE round focused on implementing the data processing and math behind ML systems. Expect 2 or 3 challenges in one session, each requiring an efficient implementation of a core ML operation.
The round shares its name with the AI-enabled coding interview on generalist loops, so ask your recruiter which format you'll face and whether an AI assistant is available. Practice implementing linear algebra and nearest-neighbor operations from scratch, with attention to sparse inputs and time complexity.
Practice from-scratch implementations of common ML building blocks, such as k-means clustering, nearest-neighbor lookups, and the forward pass of multi-head attention.
Interviewers look for:
- Efficient implementation: Whether your solution avoids unnecessary computation, such as iterating over zero entries in sparse data
- Data structure choice: How you represent sparse or high-dimensional data for the operation at hand
- Math fluency: How comfortably you apply linear algebra and statistics inside working code
- Pace: Whether you complete multiple implementations within the session
- Communication: How clearly you explain complexity and trade-offs as you code
Recently asked questions
Here are some real interview questions reported by candidates:
- Implement sparse matrix-vector multiplication efficiently.
- Implement a k-nearest neighbor search.
- Design a data structure for a sparse vector that supports efficient operations.
Behavioral interview
The Meta MLE behavioral interview is a 45-minute conversation about how you use AI to drive impact, how you keep learning, and how you partner with others, handle ambiguity, and communicate. Prepare 4-5 stories, including at least 2 where you used AI to ship something and caught an AI mistake along the way.
Expect detailed follow-ups on your most significant project, including why you chose it, its impact, and how you measured that impact. Build your story bank around these current focus areas, even if your prep materials list older ones.
Conflict and disagreement questions may get fewer follow-ups than the project discussion. Tell each conflict story end to end, including the outcome, so it’s complete without prompting.
At E6, a weak behavioral round can lead to a down-leveled offer. Prepare stories that match the scope of the level you’re targeting.
Interviewers look for:
- AI-driven impact: How you’ve used AI tools to deliver measurable results
- Continuous AI learning: How you keep up with AI tools and apply what you learn
- Partnering: How you work across teams and resolve conflict with peers and managers, including times you were wrong
- Embracing ambiguity: How you move work forward when priorities shift or direction is unclear
- Communicating effectively: How clearly you explain decisions and trade-offs to different audiences
- Measured impact: How you quantify results and tie them to business outcomes
Recently asked questions
Here are some real interview questions reported by candidates:
- Tell me about a time you were under schedule pressure to deliver something you'd committed to. What did you do?
- Tell me about a time you received a less-than-stellar review from your manager. How did you grow from it?
- How do you push development forward when you don't have explicit support from the rest of your team?
- Tell me about a time you had to pivot unexpectedly. How did you handle not having control over the direction of your work?
- Tell me about a time you worked independently.
- How have you used AI to increase your impact?
- How do you safely use AI in your day-to-day work?
Meta ML system design interview
The ML system design interview in Meta’s MLE loop is a roughly 45-minute, open-ended round where you design a production ML system end to end, most often a recommendation or ranking system. Interviewers may cut a section short and direct you to the next stage, and some give little feedback as you work.
Set your structure at the start and move through it without waiting for cues. Define the problem, align the business objective with the ML objective, and cover data and features at a high level, then save the most time for model choice and evaluation.
Keep theory and math derivations out of your answer unless the interviewer asks for them. Extra depth on one stage reduces the time left for evaluation, which interviewers expect you to cover.
Some ML system design rounds now focus on generative AI or LLM-based systems. Prepare at least one LLM design, such as a retrieval-augmented generation (RAG) feature, alongside your ranking designs.
Interviewers look for:
- Problem framing: How you define the problem and align the business objective with the ML objective
- Modeling choices: How you select and justify the model and training approach
- Data and features: How you approach data collection and feature engineering at a high level
- Evaluation and deployment: How you choose online and offline metrics that tie back to the business objective and plan to deploy and monitor the model
- Decision rationale: Whether every design decision comes with a clear explanation
Recently asked questions
Here are some real interview questions reported by candidates:
- Design a recommendation system that ranks locations in a feed for a user who just arrived in a new city. How would you featurize the data that decides which locations rank highest?
- Design an Instagram ranking model.
- Design the recommendation and ranking system for Instagram Reels.
- Design a system that detects harmful content.
- Design comments ranking on Facebook posts.
AI system design and system design interviews (specialized tracks)
Specialized Meta MLE loops, such as computer vision, can include 2 design interviews of about 45 minutes each: an AI system design round and a full system design round.
The AI system design round, also called ML modeling, is a back-and-forth on a problem that often comes from the target team’s domain. Prepare to defend the modeling trade-offs behind each idea and adjust when the interviewer pushes back.
The system design round asks you to design a complete product system that balances the algorithm, data collection, and production handling. Practice designs where the ML component is one part of a larger system, budgeting time for all 3 areas and keeping algorithm details brief unless the interviewer asks for them.
Interviewers look for:
- Modeling approach: How you frame a domain problem as an ML task and choose a suitable model
- Iteration: How you refine your approach when the interviewer questions it
- Data collection: How you plan to gather and maintain the data the system depends on
- Production handling: How you run the algorithm reliably at production scale
- System trade-offs: How you balance algorithm quality, data collection, and production constraints
Recently asked questions
Here are some real interview questions reported by candidates:
- Estimate depth from a single image.
- Design a system like Google Street View.
- Estimate how much storage Google Maps Street View requires.
How to prepare for the Meta MLE interview
- Confirm your track early: Ask your recruiter whether your loop includes AI-enabled coding or ML coding, and whether you'll face one design round or 2. The answer decides how you split prep time between coding and design.
- Practice coding at Meta's pace: Solve 2 challenges back to back in 40 minutes, add a third to some sessions, and review heap-based patterns like running top-k.
- Prepare every project to the next layer of detail: Know the motivation, assumptions, trade-offs, limitations, and validation behind each project on your resume. Project questions appear in the pre-screen, behavioral, and design rounds, and follow-ups can go several layers deep.
- Think out loud in every round: Narrate your reasoning, ask clarifying questions, and keep a visible structure when a question has no obvious answer. Interviewers across the loop evaluate how clearly you communicate under time pressure.
- Review ML fundamentals at the application level: Prepare to apply core ML concepts to concrete implementation and design choices. Expect questions that connect a concept to implementation details within the same answer.
- Practice in both coding environments: Write standard solutions in a plain editor without autocomplete, then practice separate sessions in an editor with a built-in AI assistant. Meta’s loop can include both formats, and switching between them takes practice.
- Start ML design prep as early as coding prep: Meta’s ML design prompts center on recommendation and ranking systems. Practice recommendation and ranking designs first, then expand to other ML systems.
- Practice with mock interviews: Run timed coding and ML design sessions with peer or AI mock interviews to build pace. For targeted feedback on ML design and behavioral answers, book time with an expert coach.
Additional resources
- Machine Learning Engineer Interview Course
- ML System Design Interview Course
- ML Coding Interview Course
- Software Engineering Interview Course
- Engineering Behavioral Interview Course
- Meta MLE interview questions
- Meta interview experiences
- Get a Job at Meta: Interview Process and Top Questions
FAQs about the Meta MLE interview
What is the AI coding interview at Meta?
Meta uses "AI coding" for 2 different interviews, depending on the MLE track. Generalist loops include a 60-minute AI-enabled coding round where you debug and extend a multi-file codebase with a built-in AI assistant. Specialized loops, such as computer vision, can include an ML coding round focused on implementing operations like sparse matrix-vector multiplication and k-nearest neighbor search.
How many rounds are in the Meta MLE final loop?
The Meta MLE final loop has 4-5 interviews, depending on your track. Generalist loops typically include standard coding, AI-enabled coding, ML system design, and behavioral rounds. Specialized loops can swap AI-enabled coding for ML coding and add a second design round, and E6 candidates often get 2 ML system design rounds.
Is the Meta MLE interview team-specific?
The Meta MLE interview process is team-independent through the final loop, and team matching happens after you pass. The track you apply to, such as generalist MLE or computer vision, still determines which coding and design rounds you take.
How long does the Meta MLE interview process take?
Expect roughly 4-8 weeks from first contact to a hiring decision, with results often arriving within a week of your loop. Team matching can add several more weeks, depending on location and headcount. Share any competing offer deadlines with your recruiter early, and follow up directly if you haven't heard back within 2 weeks of your loop.
How much does a Meta machine learning engineer make?
Here are the reported compensation ranges by level for Meta machine learning engineers, according to Levels.fyi:
- E3 (Entry-level MLE): ~$187K
- E4 (MLE): ~$310K
- E5 (Senior MLE): ~$458K
- E6 (Staff MLE): ~$782K
Stock makes up most of total pay at E6, where it’s nearly double base salary. Base salary rises from $138K at E3 to $261K at E6.
Want to ace your Meta interviews?
Try our Machine Learning Engineer interview course for mock interviews, frameworks, and more.
Start course for free →Get 1-1 coaching