What's the difference between 'product engineer' and 'infra engineer' tracks at Meta?

Both can reach E5+. Product engineers ship user-visible features (Feed, Reels, Marketplace). Infra engineers work on platforms below products (Tupperware, Hydra, ML training infra). The interview is the same; team matching at offer time determines track. Many engineers move between tracks during their tenure. The 'infra' specialty often pays slightly more at staff+ levels; the 'product' track has more career mobility into management roles.

Is Meta really 3 days/week in office?

Yes for most US SWE roles since the September 2023 return. Hub-specific exceptions exist — some Reality Labs research roles in Burlingame are 5 days; some engineers grandfathered into remote roles still exist but are dwindling. Most candidates interview for hybrid; remote roles are explicitly noted on the posting. Reference: metacareers.com job pages show in-office days per posting.

Does Meta interview differently for ML engineers?

Yes. ML engineering rounds include an ML system-design round and an ML-coding round (usually a model-implementation question in PyTorch). The general SWE rounds (algorithmic coding, system design, behavioral) are still part of the loop. ML candidates typically have a stronger ML focus on top of the standard SWE bar. Reference: metacareers.com/ml-prep.

What's the bar for Meta E5 (senior) vs E4 (mid)?

From Meta's published leveling and Hello Interview's analysis: E4 owns features end-to-end with a senior nearby. E5 owns technical direction for a feature area, drives cross-team decisions, and mentors E4s. The interview bar at E5 expects fluent system-design articulation, ability to drive trade-off conversations independently, and demonstrated behavioral evidence of past senior-scope work. Borderline candidates often get E4 with a re-interview path to E5 in 12-18 months.

Does Meta sponsor visas?

Yes, broadly. Meta is one of the larger H-1B sponsors in tech and supports STEM OPT, O-1, and EU equivalents. Per metacareers.com immigration guide, sponsorship varies by role and country; verify with the recruiter early. Meta has historically been more visa-friendly than Apple or Microsoft in tech.

Do I need to know React or Hack before applying?

No. Meta's interview accepts Python, Java, C++, JavaScript, Go on Coderpad. Specific language knowledge isn't tested in the loop. Post-offer, expect to ramp into Hack (web roles) or your team's language. Engineers from polyglot backgrounds typically ramp fastest; deep specialists in one non-Meta language take longer.

What's the relationship between Meta's published prep guide and what actually happens in the interview?

The published guide (metacareers.com/swe-prep-onsite) is accurate and useful but generic. It tells you the rounds, the format, sample problem categories. It does not tell you the difficulty progression, the depth of behavioral probing, or the specific system-design problems. Supplement with Hello Interview's senior-bar walkthroughs and Blind / teamblind.com's recent on-site reports for current detail.

How does Meta's Performance Summary Cycle (PSC) actually work?

Twice yearly (H1 + H2), each engineer is rated on impact + behaviors. Ratings range from 'Below Expectations' (rare, often performance-improvement plan) through 'Meets Expectations' (most common) to 'Greatly Exceeds Expectations' (top tier). Ratings drive bonus, equity refresher, and promotion eligibility. The variance is real — 'Greatly Exceeds' refreshers can be 2-3x the 'Meets' refresher. Reference: candidate reports on Blind document the structure; Meta does not publish PSC details externally.

Software Engineer Hub

Software Engineer at Meta (2026)

By Blake Crosley · Last verified 2026-04-29

In short

Meta operates one of the largest engineering organizations in tech with E3-E8 levels for ICs (E3 = junior through E8 = principal+; the public title 'Software Engineer' covers E3-E7). The primary stack is Hack/PHP for the web monolith, React + Relay + GraphQL for client work, C++ for performance-critical systems, Python (with PyTorch) for ML, and Java/Kotlin / Swift for mobile. Senior (E5) total comp clusters $440k-$590k per levels.fyi/companies/facebook/salaries/software-engineer/levels/e5.

Key takeaways

Meta uses E3-E8 IC leveling — E5 (senior) is where most engineers spend their career; E6 (staff) is reached by a smaller subset (per Hello Interview's FAANG levels guide, hellointerview.com/blog/understanding-job-levels-at-faang-companies).
Coderpad is the standard coding-interview environment; expect 2 coding rounds plus a behavioral and (at senior+) a system design round in the on-site loop.
Meta's primary web stack is Hack (a typed dialect of PHP that Meta built and open-sourced via HHVM at hhvm.com); not Java, not Python — engineers who haven't used Hack ramp from PHP/TypeScript/Java backgrounds.
Meta's E5 senior-bar coding round expects two solid solutions in 45 minutes — including time discussing trade-offs and edge cases (per Meta's published interview prep guide at metacareers.com/swe-prep-onsite).
Returned-to-office in 2023; most US SWE roles are 3 days/week in office at Menlo Park, NYC, Seattle, or Bellevue per metacareers.com.
Meta's engineering blog (engineering.fb.com) has substantial published content on Hydra, Tupperware, Cubrid, and the Llama AI infrastructure — required pre-interview reading for senior+ candidates.

Where Meta SWEs work — products and infra

From metacareers.com/jobs (verified 2026-04-27), the major engineering surfaces:

Family of Apps (FoA). Facebook (web + mobile), Instagram, WhatsApp, Threads, Messenger. The historical core; tens of thousands of engineers across these products.
Reality Labs. Quest hardware/OS, Horizon Worlds, smart glasses (Ray-Ban Meta), AR research. C++/Unity/Unreal heavy; specialty graphics and embedded skill sets.
AI infrastructure (FAIR + Generative AI). Llama models, AI Studio, Meta AI assistant, Aria. Python/PyTorch primarily, with a heavy distributed-systems layer (PyTorch Distributed, Megatron-style training infrastructure).
Ads & Business Platform. Ad targeting, attribution, billing, advertiser-facing tools. Meta's revenue-bearing surface; large engineering footprint.
Infrastructure. Compiler infra, networking, datacenter, Tupperware (container scheduler), Hydra (config), Cubrid (storage). Documented in engineering.fb.com.
Trust & Safety / Integrity. Content moderation tooling, ML for harmful-content detection, abuse mitigation.
Internal platforms. Phabricator (now legacy), internal IDE, deploy systems, mobile build infrastructure. The 'engineering effectiveness' org has hundreds of engineers.

Meta SWE roles are typically 'product engineer' (E3-E5 lean toward this) or 'infra engineer' tracks; specialization within each is by team. Most candidates interview generic and bootcamp into a team after offer.

The interview process: published guide + candidate reports

Meta is unusually transparent about its SWE interview process — the official prep guide is at metacareers.com/swe-prep-onsite. Synthesized with Glassdoor / Blind / candidate reports, the shape:

Recruiter screen (30 min).
Coding screen (45 min, Coderpad). One or two algorithmic problems. Meta's published guidance: 'two solutions in 45 minutes including discussion'. Don't slow down to perfect one; communicate trade-offs.
On-site (4-5 rounds, Meta calls them 'Ninja' for full SWE loop):

Coding rounds (2). Expect medium-tier LeetCode equivalents — graph, DP, sliding window, trees. Meta's culture: clean working code > over-optimized code. Walk through your reasoning. Edge cases matter.
System design round (E5+ only). Standard format: requirements → estimation → high-level design → deep dive → trade-offs. Meta's interviewers tend to push on consistency and partition-tolerance — DDIA chapter 5 and chapter 9 are the canonical references.
Behavioral / 'Jedi' round. Meta's behavioral round is rigorous — they probe past projects deeply, looking for impact, collaboration, conflict resolution. Star/SAR (Situation-Action-Result) framing helps. Specific numbers help more.

Meta runs the same on-site for E3-E7. Leveling decisions happen post-interview based on the bar each interviewer felt the candidate met. Common outcome: candidate interviewed for E5 but is offered E4 if borderline; some negotiate up with prior leveling evidence.

Compensation, sourced

Meta publishes US salary ranges per pay-transparency laws on individual job postings. levels.fyi aggregated:

E3 (junior, ~0-2 yrs): ~$170k base, $245k-$330k total per levels.fyi/companies/facebook/salaries/software-engineer/levels/e3.
E4 (mid, ~2-5 yrs): ~$210k base, $310k-$425k total.
E5 (senior, ~5-8 yrs): ~$240k base, $440k-$590k total. levels.fyi/companies/facebook/salaries/software-engineer/levels/e5.
E6 (staff, ~8-12 yrs): ~$300k base, $640k-$870k total.
E7 (senior staff): ~$340k base, $810k-$1.2M+ total.
E8 (principal): ~$400k+ base, $1M-$1.8M+ total.

Stock-vesting note: Meta moved to a tiered RSU vesting structure in 2014 (still in effect): 25/25/25/25 over 4 years for new grants, with refresher grants stacked annually. The 'cliff effect' year-1 means new hires see stock comp ramp meaningfully in years 2-4 as refreshers stack.

Meta-specific bonus: Meta's Performance Summary Cycle (PSC) results determine annual bonus and equity refresher. Top performers (Greatly Exceeds Expectations) get materially larger refreshers; the variance year-over-year is real and intentional.

Meta's stack and tooling: what to expect day one

Meta's engineering stack is unusual; candidates from non-FAANG backgrounds often need a 4-8 week ramp.

Hack (a typed PHP dialect): the dominant web language. Meta open-sourced HHVM (the runtime) and Hack (the language) at hhvm.com. Most product code is Hack. Engineers from Java/TypeScript backgrounds typically ramp in 2-4 weeks.
React + Relay + GraphQL: Meta invented React (2013) and Relay (2015); they remain the canonical client stack. GraphQL came from Meta in 2015; Meta's internal usage is more sophisticated than most public examples.
C++ for performance-critical: news feed ranking, Hydra (config service), graph storage. C++ is the senior-bar IC track for performance engineering.
Python + PyTorch for ML: Meta open-sourced PyTorch (2016) and most Llama models. Most ML engineering is Python with the FBLearner ML platform.
Mobile: Java/Kotlin (Android), Swift (iOS), React Native (cross-platform for some surfaces).
Internal tooling: Phabricator (legacy), Mercurial (the SCM, not Git — Meta's monorepo uses sapling.dev, formerly EdenSCM). Internal IDE 'Bento' for notebooks; FBLearner for ML pipelines.

The monorepo: Meta runs one of the largest monorepos in tech (~100M+ lines across Hack, C++, Python, Java). Sapling (sapling-scm.com, open-sourced 2022) is the version control. The build system is Buck2 (open-sourced 2023). Day-one ramp involves learning these tools.

Frequently asked questions

What's the difference between 'product engineer' and 'infra engineer' tracks at Meta?: Both can reach E5+. Product engineers ship user-visible features (Feed, Reels, Marketplace). Infra engineers work on platforms below products (Tupperware, Hydra, ML training infra). The interview is the same; team matching at offer time determines track. Many engineers move between tracks during their tenure. The 'infra' specialty often pays slightly more at staff+ levels; the 'product' track has more career mobility into management roles.
Is Meta really 3 days/week in office?: Yes for most US SWE roles since the September 2023 return. Hub-specific exceptions exist — some Reality Labs research roles in Burlingame are 5 days; some engineers grandfathered into remote roles still exist but are dwindling. Most candidates interview for hybrid; remote roles are explicitly noted on the posting. Reference: metacareers.com job pages show in-office days per posting.
Does Meta interview differently for ML engineers?: Yes. ML engineering rounds include an ML system-design round and an ML-coding round (usually a model-implementation question in PyTorch). The general SWE rounds (algorithmic coding, system design, behavioral) are still part of the loop. ML candidates typically have a stronger ML focus on top of the standard SWE bar. Reference: metacareers.com/ml-prep.
What's the bar for Meta E5 (senior) vs E4 (mid)?: From Meta's published leveling and Hello Interview's analysis: E4 owns features end-to-end with a senior nearby. E5 owns technical direction for a feature area, drives cross-team decisions, and mentors E4s. The interview bar at E5 expects fluent system-design articulation, ability to drive trade-off conversations independently, and demonstrated behavioral evidence of past senior-scope work. Borderline candidates often get E4 with a re-interview path to E5 in 12-18 months.
Does Meta sponsor visas?: Yes, broadly. Meta is one of the larger H-1B sponsors in tech and supports STEM OPT, O-1, and EU equivalents. Per metacareers.com immigration guide, sponsorship varies by role and country; verify with the recruiter early. Meta has historically been more visa-friendly than Apple or Microsoft in tech.
Do I need to know React or Hack before applying?: No. Meta's interview accepts Python, Java, C++, JavaScript, Go on Coderpad. Specific language knowledge isn't tested in the loop. Post-offer, expect to ramp into Hack (web roles) or your team's language. Engineers from polyglot backgrounds typically ramp fastest; deep specialists in one non-Meta language take longer.
What's the relationship between Meta's published prep guide and what actually happens in the interview?: The published guide (metacareers.com/swe-prep-onsite) is accurate and useful but generic. It tells you the rounds, the format, sample problem categories. It does not tell you the difficulty progression, the depth of behavioral probing, or the specific system-design problems. Supplement with Hello Interview's senior-bar walkthroughs and Blind / teamblind.com's recent on-site reports for current detail.
How does Meta's Performance Summary Cycle (PSC) actually work?: Twice yearly (H1 + H2), each engineer is rated on impact + behaviors. Ratings range from 'Below Expectations' (rare, often performance-improvement plan) through 'Meets Expectations' (most common) to 'Greatly Exceeds Expectations' (top tier). Ratings drive bonus, equity refresher, and promotion eligibility. The variance is real — 'Greatly Exceeds' refreshers can be 2-3x the 'Meets' refresher. Reference: candidate reports on Blind document the structure; Meta does not publish PSC details externally.

Sources

About the author. Blake Crosley founded ResumeGeni and writes about product design, hiring technology, and ATS optimization. More writing at blakecrosley.com.