The Meta System Design Interview: Product Architecture vs System Design, E4 to E6

System Design is the round for infrastructure roles (Software Engineer, Infrastructure). It leans toward the back end: storage, queues, partitions, what happens when a machine dies. Product Architecture is the round for product roles (Software Engineer, Product). It leans toward the shape of a product: the API, the data model, how a client and a server talk, what a user sees while they wait.

A rule of thumb that coaching guides attribute to Meta: if the whole product could be built from off-the-shelf cloud parts, it is a Product Architecture question. The same guides admit the line is blurry, and quote an internal view that there is a lot of ambiguity about the difference. In practice the same prompt can show up in either round. Hello Interview reported that about 15% of its candidates were asked to design only the screens and APIs, usually for a news feed.

Ask your recruiter which track you are on, then lean your practice toward it. A System Design candidate should be ready to go below the database. A Product Architecture candidate should be ready to write the API calls in full, including pagination and what the client caches.

Meta’s candidate page says the design interview is 45 minutes and that the interviewer usually keeps the last five for your questions. The page is dated October 6, 2022, and still said so on October 24, 2026. Coaching guides report one design round at E4 and E5, and usually two at E6. Many loops also have a design question in the technical screen.

The usable time is shorter than the booking. In a recorded Meta-style practice round on interviewing.io, the interviewer tells the candidate that about 35 of the 45 minutes are for the problem. Plan for 35 minutes: a few for requirements, a few for numbers, a working design by about minute 25, then the deep dives.

Both tracks are scored on the same four axes, as Hello Interview’s Meta guides describe them. No weights are published.

• Problem navigationFinding the requirements that matter and the part of the problem that is hard, then spending your time there.
• Solution designAn architecture that works and scales, with the trade-offs between speed, scale, upkeep and cost said out loud.
• Technical excellenceDepth in the pieces you chose: what the cache key is, what the lock holds, what happens when a node dies.
• Technical communicationExplaining each decision and its reason so the interviewer can follow, and write it down.

Meta’s own page says the same things in plainer terms. It wants clarifying questions that still drive toward a solution, a plan broken into parts, focus on the interesting and hard problems, movement between the big picture and the detail, and trade-offs weighed out loud. It also says that spotting and working around typical problems is part of the signal.

This is changing, so check with your recruiter. On October 24, 2026, Meta’s hiring process page said design interviews run in CoderPad with Mermaid Markdown, a text format that draws boxes and arrows from lines like client --> api. The page said the environment has a built-in AI assistant and that only the tools inside it may be used.

Coaching guides read the same day still described Excalidraw, a free drawing board, as the standard tool for both tracks, and an older Meta post from 2022 let candidates share any drawing tool they liked. If you get CoderPad with Mermaid, practise writing a diagram as text before the day: it is slower than dragging boxes, and a typo can hide half your design.

Meta’s page gives its own examples: the back end of a messaging system, an email server, and the API for a rich document editor. Candidate reports compiled by coaching sites point to a short list that keeps coming back:

• Design LeetCode, reported as the most asked on both tracksPractise this one
• Design an ad click aggregator, System Design trackPractise this one
• Design a game leaderboard, System Design trackPractise this one
• Design Ticketmaster, asked at E4, E5 and E6Practise this one
• Design a top songs widget, Product Architecture trackPractise this one
• Design an online auction, Product Architecture trackPractise this one

Instagram and the news feed come up often too, and a Meta commenter on Blind called Ticketmaster a favourite. What each of these needs, and the deep dives interviewers push toward, is in 62 system design interview questions, and who asks them.

Coaching guides call the design round the one that carries the most weight for your level, along with the behavioral round. At E4 it is described as crucial for setting the level at all. The level guides we read on October 24, 2026 describe what changes from one level to the next. Applied to one question, LeetCode, it looks like this:

The E6 guide says being placed at E5 instead is common, often over depth in the design rounds. Coaching guides name two more patterns that cost a senior candidate a level: building more than the problem needs, and waiting for the interviewer to lead. At E4, some steering from the interviewer is expected and costs little.

Candidates describe Meta interviewers pressing for throughput maths and cache invalidation, and scaling the problem up again and again. One candidate in Hello Interview’s E5 guide redesigned their sharding three times as the numbers grew. Expect "what if there are 100 times more users?" more than once.

The same guide says Meta interviewers are trained to stay neutral, so a flat face tells you nothing. Do not wait for a nod before you move on. Silence hurts you more than a wrong turn you notice and fix. And when the interviewer narrows the scope, for example to the API alone, follow at once.

No company publishes a time split, but coaching guides agree on the shape. For a Meta round with about 35 usable minutes, a plan that leaves room for depth looks like this: five minutes on requirements, with three things the system must do and a few targets with numbers. Three minutes on the numbers that will pick a component. Five minutes on the core entities and the API. About twelve minutes on a design with a path for every requirement. Then ten minutes on one or two deep dives.

Shift the weight with the track. In Product Architecture, spend longer on the API and the data model: the calls in full, how a list pages, what the client keeps. In System Design, get through the API quickly and save the time for what happens under load. Meta’s own example shows the kind of follow-up to expect: for a messaging back end, how many messages, how fast, how much storage, and whether new messages are pushed to the phone or pulled.

Some candidates now report a third kind of design round: a 45-minute AI-enabled interview where an assistant inside the tool helps with ideas, research and brainstorming. This comes from a Blind thread, not from Meta, so treat it as a report. A Meta employee replying there suggested preparing retrieval pipelines for documents (RAG), agents that call tools, how to evaluate the answers, and safety guardrails. Meta’s hiring page does confirm the built-in assistant.

After the round, the interviewer writes a summary of what you did, with a recommendation from low to high confidence and a read of your level. A committee that never saw your board reads that summary. A rule of thumb in the community, and not an official one, is that coding decides whether you are hired and design decides at what level.

So give the interviewer moments they can write down: three requirements with numbers before the first box, a number that picks a component, the bottleneck named before the fix, and a trade-off in one sentence. For how that sheet fills up minute by minute, read what the interviewer writes down.

Star Interview runs a 45-minute practice system design interview on a shared whiteboard. Choose Meta and your level, and your AI interviewer picks a question for that level, drawing on the questions Meta candidates report where it can, then answers your questions with numbers, runs two deep dives and scales the problem up partway. Your report is scored on the same kind of axes, with the level your design read as.