Mercor interview questions reported by the community

TRPCClientError

Debug CSV Serialization Performance and Trailing Delimiters

Permutation-Based Sorting

Review and Improve a CSV Builder

Currency Exchange Rate Lookup

Thread-Safe Compute-Once Cache

Shortest Path in a Grid with Obstacles Elimination

Implement and Train an Autoregressive Transformer Language Model on Shakespeare

Top K Elements with a Deterministically Faulty Comparator

Expected Network Size After N Days (Probabilistic Growth with Caps)

Rank Top-K Users by Network Reach and Flow Centrality

Implement a Referral Network (Graph Design)

Build an LLM Evaluation Harness with Ollama Integration

Amazon's bar leans on Leadership Principles and scenario thinking more than raw algorithm difficulty. Coding rounds skew toward OOD and under-specified prompts that reward clarification skill, and a GenAI-usage behavioral question appears in nearly every modern loop. Bar Raiser independence and writing-heavy work samples shape the senior-level signal.

Amazon

Meta sits mid-transition between two interview loops in 2026. The legacy track still drives most candidates through CodeSignal-style OAs and tightly time-boxed coding rounds, while a new AI-Native loop layers in AI-assisted coding, AI-Enabled system design, and an AI-oriented behavioral round on top. Which track you land in depends on org and recruiter, not candidate preference.

Google

ByteDance / TikTok runs a strict one-round-at-a-time funnel with sharp org-by-org variance in bar, language, and round mix. Coding rounds skew toward classic LeetCode-medium with one self-built test case; MLE and SRE loops front-load 30+ minutes of oral 八股 before any code is written. Interviewer professionalism varies — strong candidates drive the round structurally rather than waiting to be assessed.

Bytedance

Uber's loop leans heavily on Uber Eats and rider/driver marketplace problems — Cart pricing, geo heatmaps, surge, and restaurant recommendation surface in nearly every onsite at L5+. Coding rounds are mostly tagged LeetCode patterns, but Hack2Hire OAs ship a small rotating set of original problems that rarely appear elsewhere. Coding bar is the gate; system design quality drives leveling between L5a / L5b / Senior / Staff.

Uber

Microsoft runs two parallel hiring tracks: the standardized Hiring Event (HE) loop covering Azure, Cloud, Copilot, and Office orgs, and Mustafa Suleyman's MAI org with its own recruiter pool and a distinctly OpenAI-flavored coding rotation. The HE bar is consistent — four 60-minute rounds back-to-back on a fixed weekday — with medium-difficulty algorithms and one system-design slot; MAI loops add project deep-dive depth and pull problems from an in-memory-DB / beam-search / streaming-decoder pool that does not overlap with classic LeetCode prep.

Microsoft

OpenAI pairs an unusually high raw-coding bar with explicit no-AI-tools rules during most interviews — a deliberate contrast with the product. Phone screens often run a single multi-part simulation for the full hour, and senior loops add deep ML and research-grounded system-design rounds that probe mechanism, not vocabulary.

Openai

Snowflake's loop is small and bank-driven: a recruiter screen, two back-to-back coding phone screens, then a virtual onsite that mixes coding, system design, and behavioral. The same handful of coding families (multi-source BFS on a grid, course-schedule variants, tree-node-deletion height problems, transactional KV stores) cycle through most candidates, so recognition speed matters more than novel problem solving.

Snowflake

Apple interviews are unusually team-owned: the same company can run classic LeetCode screens, MLE case studies, hardware verification drills, project deep-dives, and system design loops with little advance disclosure. The strongest pattern is role fit: interviewers probe whether your prior work maps directly to the team's product surface, data domain, and engineering stack, while still expecting clean code, concise trade-off reasoning, and polished behavioral stories.

Apple

Stripe favors long realistic prompts in a full IDE — integration, debugging, refactor — over LeetCode-style algorithms. System design is grounded in payments and regulatory reality, and behavioral signals weigh ownership and written communication as heavily as raw technical depth.

Stripe

Databricks interviews are unusually bank-driven: candidates who know the recurring prompt families often recognize the round within the first few minutes. The loop mixes algorithmic coding with low-level systems and product-grounded design, then puts real weight on behavioral, hiring-committee, and reference signals after the technical rounds.

Databricks

Pinterest's loop leans hard on a recurring custom prompt library — escape-room state-tracking, trie-backed access control, ads/log aggregation, splitwise-style settlement — over canonical LeetCode. ML system design is split into two distinct rounds ("ML Practitioner" and "ML System Design") that effectively cover the same recommender / retrieval / ranking surface from two angles, with embedding-heavy product framing throughout. Fully-remote, front-loaded compensation, and long team-match windows define the experience as much as the technical bar.

Pinterest

DoorDash's loop is dominated by a small, slow-moving library of original Code Craft prompts (Dasher Pay, Bootstrap API, Round Robin debugging) plus a handful of recurring system-design favorites (3-day Donation, Notification, Food Review). Coding bar is the gate; system design correctness and behavioral consistency drive leveling. The bar swings sharply with the interviewer — multiple candidates report bug-free coding rounds that still ended in rejection — so prepping the canonical question set cold matters more than chasing breadth.

DoorDash

Citadel and Citadel Securities run two distinct loops sharing one brand. Citsec (the market maker) hires for HFT software, quant research and quant dev; its phone screens lean on small, sharp coding prompts (merge-K, ring buffers, stock-DP ladders) and the loop is famously single-fail — any weak round and the next interview is cancelled within a day. Citadel (the hedge fund) and its sub-orgs (GQS, EQR, NXT) lean harder on probability brainteasers, deep resume drilling and a multi-week team-match stage where coding bar drops but cultural and PM-style fit dominates the offer decision.

Citadel

Netflix's current engineering loop is high-signal and repeat-heavy: Ads candidates see the same topo / frequency-cap / ads data-modeling trio, Member-Commerce-Games candidates see homepage dedupe and duplicate-detection variants, and Infra / Data Platform candidates see cache, latency, and concurrency rounds. The technical prompts are usually solvable, but the grading is communication-heavy: candidates are expected to clarify vague requirements, write tests, explain trade-offs, and connect their past work tightly to the target org.

Netflix

LinkedIn's loop pivoted heavily in 2025 — Staff candidates now sit a packed two-round phone screen (coding + project retro) before a five-round onsite that includes an AI-Coding session in CoderPad's assistant pane. Behavioural depth is the dominant downlevel axis; the same coding bar that passes at Senior frequently caps a Staff loop with a downlevel offer.

LinkedIn

Roblox interviews lean heavily on a small, stable repertoire of coding and system design prompts drawn from the company's own product surface — rate limiting, like-button counters, matchmaking queues, delayed payments. The bar is high but predictable: interviewers want bug-free implementations with crisp boundary handling and design discussions that ground every trade-off in real platform scale (millions of concurrent players, hot creator experiences, cross-region latency). HM and director rounds heavily probe past experience with comparable traffic and scope.

Roblox

Capital One's engineering loop is unusually predictable: a long CodeSignal OA gates everyone, then a single Power Day stitches four 45-60 minute rounds — coding, system design, case study, and behavioral — back to back. The technical bar is moderate but the case study and BQ rounds carry real weight, especially for MLE / Applied Research / Data Science where the business-context framing of a model decision is the differentiating signal.

Capital One

IBM's recent technical screens are dominated by auto-triggered HackerRank online assessments for entry-level, intern, full-stack, backend, AI, and .NET roles. The question mix is mostly easy-to-medium data structures, sliding-window, interval, graph-connectivity, and small frontend/CSS tasks, while Red Hat / AI-infrastructure loops add live coding around inference kernels, Kubernetes controllers, and DevOps fundamentals.

Anthropic runs an unusually small, transparent technical question bank — the recruiter typically names which prompt family you'll see days before the round, so candidates can target preparation precisely instead of guessing. Bars are calibrated for problem-solving velocity and follow-up depth rather than novelty; offers are decided by hiring committee + reference check, and the culture round on AI-safety alignment is the single most common rejection point even for candidates who clear every technical stage.

Anthropic

Oracle's loops split along two clear axes: OCI (Cloud Infrastructure) runs a fairly standard LC + system-design loop with heavy Indian interviewer rotation, while OHAI (Oracle Health, the post-Cerner business unit) leans into OOD/class-design problems built around healthcare scenarios. Coding bars are mostly easy-to-medium, but interviewer variance and post-loop down-leveling / ghosting are unusually common — the loop result often determines the eventual level, not just the yes/no.

Oracle

Bloomberg's loop is a coding-heavy gauntlet: three back-to-back technical rounds each pack two LeetCode-style problems with little hand-holding, plus an HR screen and an engineering manager round that anchors on a single deep project story. The coding bar emphasizes pristine communication — dry-running, edge cases, and complexity discussion are graded as heavily as the solution itself, and interviewers routinely push for second-best alternatives even after a working answer. New-grad and tech-screen verdicts often arrive within hours, while jump-shift onsites favor candidates who can explain past trade-offs without prompting.

Bloomberg

Salesforce's loop is bank-driven: a small set of coding prompts (LFU Cache, Flatten JSON, linked-list dedup, frontend curry + traffic light, OA HackerRank classics) and one signature system design (Coffee Ordering, with its SQL follow-ups) cover the bulk of recent rounds. Down-leveling at offer time is common — the HM screen and behavioural depth materially shape the final level, often more than coding correctness alone.

Salesforce

Goldman Sachs interview loops lean heavily on traditional LeetCode-style coding — the firm publishes its own tagged problem list on LeetCode, and the OA / phone screens recycle from a stable bank that maps closely to it. Behavioral signal is concentrated in a Hirevue video screen with tight 30-second prep windows, and quant-flavored desks (GSAM, Quant Engineering) layer probability and statistics drills on top of the standard coding bar.

Goldman Sachs

Airbnb's loop is famously consistent and famously slow. The coding pool is small and reused — a candidate who has drilled the canonical questions (split-stay, terrain water-drop, KV store, code review) walks in advantaged. Differentiation happens in the code-review round, the experience deep-dive, and the standalone core-values round, which is calibrated separately from the rest of the loop and serves as a real gate. Expect strong technical performance to land in a multi-week team-match queue rather than a fast offer.

Airbnb

Coinbase runs a long, gated pipeline — AI recruiter screen, a cognitive aptitude test, a culture/personality questionnaire, and a CodeSignal coding OA all precede the recruiter call, and team match still happens before the 4-round VO. Coding rounds bias toward multi-level CodeSignal-style problems (banking system, in-memory database, mining-block knapsack, crypto order management) over canonical LeetCode; bring your own scaffolding because interviewers often hand you a blank file with no class structure or tests. AI-assisted coding is now an explicit round on its own.

Coinbase

Rippling's engineering loop is unusually concentrated: a small set of OOD coding prompts, two data-heavy system-design prompts, and a practical web API round account for most SWE interviews. The bar is driven by speed through multi-part prompts, clean extensible design, and whether the candidate can collaborate with AI tools without losing ownership of the solution.

rippling

Snap interviews are highly interviewer-dependent: candidates report classic graph and cache problems, open-ended backend coding, product-heavy system design, and MLE loops that dig into recommendation systems, ranking, fundamentals, and project judgment. The safest preparation strategy is to be fluent in BFS/DFS, cache/data-structure implementation, feed and ads design, and concise behavioral stories around innovation, ambiguity, and execution.

Snapchat

Waymo loops mix nominally LeetCode-style coding with deliberately ambiguous, autonomy-flavored prompts and a signature 'data fluency' round that's hard to drill for. Coding bars are mid-hard but interviewers frequently swap in original problems framed around self-driving scenarios; ML and senior loops add a behavior-prediction-style ML system design round on top.

Waymo

JPMorgan Chase's engineering loop is predictable but operationally uneven: a light HackerRank-style OA is followed by a Super Day with coding, code review, system design, and behavioral rounds rotating across candidates. The technical bar skews practical: simple algorithms, review of production-ish Java code, concurrency/style critiques, and system designs around traffic, search, inventory, payments, and URL routing.

Companies

Mercor Interview Questions

Coding Problems

Mercor Featured Guides