Meta's loop is FAANG's most stable: 2× coding + 1× system design + 1× product sense + 1× BQ. Stable doesn't mean easy — every round has Meta-specific signals. This guide breaks the full onsite down with signals, answer frameworks, and a VO assist playbook.
Meta Onsite Snapshot
| Round | Format | Duration | Focus |
|---|---|---|---|
| Coding 1 | CoderPad | 45 min | 2 LC Medium |
| Coding 2 | CoderPad | 45 min | 2 LC Medium |
| System Design | Whiteboard video | 45 min | Large-scale distributed systems |
| Product Sense / PM | Video | 45 min | Growth / metrics / design |
| BQ (Jedi) | Video | 45 min | Behavioral / values |
Stage 1: Two Coding Rounds
Meta Coding Signals
- 2 LC Medium in 45 minutes: ~22 minutes per problem
- Talk before coding: walk through the approach first
- Optimize proactively: after the first pass, expect "can we go from O(n²) to O(n log n)?"
- Test proactively: give 2–3 corner cases and dry-run after finishing
Meta High-Frequency Patterns
- 2D grids / DFS / BFS (islands, shortest path)
- Sliding window + strings (LRU, longest substring)
- Binary trees / Trie (path sum, autocomplete)
- Design problems (Twitter feed, chat room)
- Graph algorithms (topological sort, union-find)
Example: LRU with Sliding Window Twist
from collections import OrderedDict
class LRU:
def __init__(self, cap):
self.cap = cap
self.cache = OrderedDict()
def get(self, key):
if key not in self.cache:
return -1
self.cache.move_to_end(key)
return self.cache[key]
def put(self, key, val):
if key in self.cache:
self.cache.move_to_end(key)
self.cache[key] = val
if len(self.cache) > self.cap:
self.cache.popitem(last=False)
Trap: move_to_end must pair with popitem(last=False). The most common bug is using a plain dict + list (O(n) removal) and timing out.
Stage 2: System Design
Meta System Design Signals
- 45 minutes: 5 min clarify + 30 min design + 10 min wrap-up
- Real-world Meta surface: News Feed, Instagram Story, Messenger, Live Comment
- Signals: data model, scalability bottleneck, consistency vs availability
10-Step Framework
- Clarify (scale numbers, read/write ratio, latency targets)
- Capacity estimation (QPS, storage)
- Data model
- High-level architecture
- API design
- Storage choices
- Caching strategy
- Async processing (queues / batches)
- Scalability bottlenecks
- Failure recovery
Meta Real Question: Design News Feed
Signals:
- Fan-out on write vs fan-out on read (fan-out on read suits power users)
- Hybrid policy: mega-users use pull, regular users use push
- Ranking: CTR + recency + diversity
- Cache: Redis with the last ~200 feed items
Stage 3: Product Sense (PM / Some SDEs)
Surface
- "Meta wants to ship feature X — how do you define success?"
- "Reels D30 retention dropped 5pp — root cause?"
- "Design a new content moderation system — how do you trade off FP vs FN?"
Framework
- Clarify: persona / window / data source
- Decompose: north-star → driver tree → input metrics
- Hypothesis ranking: probability × impact
- Validation plan: data pull + A/B + qual
Stage 4: BQ (Jedi)
Meta Five Values
- Move Fast
- Focus on Long-term Impact
- Build Awesome Things
- Live in the Future
- Be Direct & Respect Your Colleagues
Surface
- "Most challenging conflict and how you handled it"
- "Proudest project + quantified impact"
- "Failed project and what you learned"
STAR + Quantified
Every story must include:
- Numbers: impact, headcount, time
- Reflection: what you'd do differently
- Values mapping: which Meta value it demonstrates
Meta Loop Timing
| Step | Median |
|---|---|
| Recruiter → phone screen | 5–10 days |
| Phone screen → onsite | 2 weeks |
| Onsite → verbal offer | 1 week |
| Full loop | 4–6 weeks |
Community pass rates: phone screen ~30%, onsite ~25%.
VO Assist Playbook
What oavoservice VO assist gives you
- 45-minute dual-problem coding sim: CoderPad with 2 LC Medium + proactive optimization drills
- System design 10-step framework: News Feed / Instagram Story / Live Comment whiteboard scripts
- Product Sense case library: 15 real Meta / Instagram / WhatsApp / Reels cases
- BQ Jedi bank + values mapping: 30 BQs with value-mapping templates
What's hard about Meta loops
Interviewers explicitly score speed + proactive optimization. We've seen candidates with correct code get weak-signaled because they never said "can we optimize?" VO assist drills the state → write → optimize → test four-step reflex on every problem.
Add WeChat Coding0201 for pricing and scope.
FAQ
Phone screen vs onsite difficulty?
Phone screen: 1 LC Medium in 45 minutes. Onsite: 2 LC Medium in 45 minutes. Onsite demands 2× the pace.
Drawing required for system design?
Strongly recommended. Excalidraw or a physical whiteboard works — focus on data flow + data model.
What if my BQ doesn't fit STAR?
Meta BQ rewards authenticity + quantification. If STAR doesn't fit, just tell the story, but always include numbers + reflection.
Cooldown after no offer?
12 months. Cross-role (SDE → ML / PE) typically uses a separate pool.
Preparing for Meta / Instagram / WhatsApp / Reality Labs?
oavoservice tracks Meta / Instagram / WhatsApp / Reality Labs / Quest interviews for SDE / PE / ML / PM. Our mentors come from live News Feed / Reels / Messenger / Ads teams and deliver 45-minute dual-problem coding sim, 10-step system design, Product Sense case library, and BQ Jedi mocks.
👉 Add WeChat: Coding0201 for the Meta full interview prep + VO assist plan.
Contact
Email: [email protected]
Telegram: @OAVOProxy