Shortest Path in a Grid with Obstacles Elimination (LeetCode 1293): Given an m x n grid where cells are 0 (empty) or 1 (obstacle), find the minimum number of steps to walk from the upper-left to the lower-right corner given you can eliminate at most k obstacles.

The key insight is BFS with a 3-tuple state (row, col, remaining_eliminations) rather than just (row, col). Each state is visited at most once, giving O(m * n * k) time and space. A strong candidate immediately recognizes that standard BFS is insufficient because the 2D position alone does not capture remaining budget, and they encode the budget as a third dimension of the visited set. Likely follow-up: what if obstacle costs are variable? Ability to augment BFS state when constraints make 2D visited sets insufficient; correct handling of the k budget across branches.

Longest Substring Without Repeating Characters (LeetCode 3): Given a string, find the length of the longest substring without repeating characters.

Classic sliding window with a hash map tracking the last seen index of each character. Advance the left pointer to max(left, last_seen[char] + 1) when a duplicate is encountered. O(n) time, O(min(n, alphabet)) space. The key insight is that the left pointer only ever moves right, so no reprocessing occurs. Follow-up: handle Unicode characters or return the actual substring. Clean two-pointer implementation, correct pointer update logic, ability to reason about edge cases (all unique, all same, empty string).

LRU Cache (LeetCode 146): Design a data structure that supports get(key) and put(key, value) in O(1) average time, evicting the least recently used entry when capacity is exceeded.

Combine a hash map (key to doubly-linked-list node) with a doubly linked list (head = MRU, tail = LRU). On get or put, unlink the node and reinsert at head; on capacity overflow, remove the tail node and its hash map entry. O(1) for both operations. A strong candidate draws the invariant explicitly before coding and uses sentinel head/tail nodes to avoid null checks. Follow-up: extend to a TTL-based or priority-based eviction policy. Correct pointer manipulation, understanding of why a doubly linked list is required (O(1) deletion), and clean encapsulation.

Design a monitoring and alerting system where users can configure alert rules on top of a metrics stream. The system should ingest time-series metrics, evaluate configurable threshold or anomaly-based rules, and send notifications when rules trigger.

Start with requirements: metrics ingestion rate, rule evaluation latency SLA, alert fanout, and deduplication window. Core components: an ingestion layer (Kafka or Pulsar for durability), a rule evaluation service (sliding-window aggregation in a stream processor like Flink), a rule store (low-latency K/V such as Redis or DynamoDB), and a notification dispatcher with deduplication. Key trade-off: push-based evaluation per metric event vs. pull-based polling on a schedule. A strong candidate addresses alert flapping suppression and at-least-once vs. exactly-once semantics. Follow-up: how would you scale to 1 million concurrent rules? Structured requirements gathering, separation of the ingestion and evaluation planes, awareness of stream processing semantics, and practical deduplication/suppression logic.

Design Instagram's photo/video feed (reported in the ByteDance Dubai interview, May 2024). Given a social graph, serve a personalized, paginated feed of posts from followed accounts with low latency.

Fan-out-on-write (push) vs. fan-out-on-read (pull) debate is the central trade-off. For most users, pre-compute a feed timeline in Redis sorted sets (fan-out-on-write at post time); for celebrity accounts with millions of followers, hybrid fan-out or pull at read time. Feed pagination via cursor tokens rather than offset. CDN for media assets. A strong candidate discusses write amplification for high-follower accounts and how to blend celebrity posts at read time. Follow-up: add algorithmic ranking instead of chronological ordering. Awareness of the fan-out trade-off, correct use of sorted sets for timeline storage, CDN strategy for media, and handling of the celebrity problem.

Design a real-time messaging app with contacts syncing, offline message storage, and media sharing (reported in the ByteDance TikTok Android SWE 2-2 Singapore interview, March 2025).

WebSockets or long-polling for real-time delivery; a message queue (Kafka) as a durability buffer; per-user inbox stored in Cassandra or DynamoDB for offline delivery. Media sharing via pre-signed S3 URLs to avoid routing binaries through the API layer. Contacts sync via delta-sync with a version vector or last-synced timestamp. Key reliability concern: ensuring exactly-once delivery using client-generated message IDs and server deduplication. Follow-up: end-to-end encryption key exchange. Correct separation of signaling vs. media paths, offline delivery via persistent inbox, idempotency of message delivery, and practical contacts sync mechanism.

What is your favorite project you have worked on, and what was your specific contribution?

Use STAR to walk through the project context, your individual ownership, the technical decisions you drove, and the measurable outcome. ByteDance rewards candidates who can articulate both the engineering depth and the business impact. Avoid narrating team accomplishments without isolating your own role. Likely follow-up: what would you do differently in hindsight? Clarity of personal ownership vs. team contribution, evidence of technical depth, and ability to quantify impact aligned with ByteDance's data-driven culture.

Tell me about the biggest challenge or conflict you faced on a project or with a teammate, and how you resolved it.

Structure the answer around the conflict root cause (technical disagreement, resource contention, or priority misalignment), the steps you took to surface and resolve it (data-driven argumentation, escalation path, compromise), and the outcome. ByteDance explicitly values being grounded and courageous, so glossing over the conflict to seem agreeable reads negatively. Follow-up: how did the resolution change the way you work with that person? Intellectual honesty about disagreement, use of data or reasoning rather than authority to resolve conflict, and reflection on what was learned.

Tell me about a time you had to learn something completely new under pressure or on a very short timeline. How did you approach it?

Connect the answer to ByteDance's Rapid Learning and Self-Iteration ByteStyle. Be specific about the knowledge gap, the structured approach taken (reading primary sources, building a toy prototype, pairing with a domain expert), and the timeline. A strong answer includes a setback or wrong assumption encountered mid-learning and how it was corrected. Follow-up: what is your general framework for onboarding into an unfamiliar codebase or technology? Demonstration of meta-learning skills, comfort with ambiguity and incomplete information, and a growth mindset consistent with Always Day 1.

ByteDance Interview Questions

ByteDance (parent of TikTok) runs a 3 to 5 round SWE loop: an online assessment (historically HackerRank, switched to CodeSignal in 2026), followed by 2 to 3 live coding rounds, a system design round for E4 and above, and a behavioral/HR round. Question difficulty skews heavily Medium to Hard on LeetCode, widely reported as harder on average than other FAANG loops. The loop is largely remote; Singapore and Mountain View offices are the main international hiring hubs. Interviewers progressively modify problems within a round, requiring real-time adaptation rather than pure recall.

17 real questions across 3 rounds

Reported via LeetCode Discuss first-hand reports: TikTok/Bytedance Interview Experience (post 7150750, 2025), TikTok/ByteDance Interview Android SWE 2-2 Singapore March 2025 (post 6802671), Bytedance interview questions Dubai Round 1 (post 5178868, May 2024), ByteDance Final Round LeetCode Discuss (post 1076352). NodeFlair blog compiled from 243 interviews. Glassdoor ByteDance SWE reviews 2024-2025. TikTok Online Assessment 2026 Lodely breakdown. freezefrancis Medium article on TikTok Singapore backend engineer interview.

Run a ByteDance mock with the AI interviewer. It asks questions like these, follows up on your answers, and tells you exactly what to fix. The only score that means anything is the one on your real answers.

Continue with Google

Keep prepping

All company interviews