Interview Experiences

Amazon Prime Video Ads Channels SDE II: Locker & Package OOD, reordering ads so one advertiser never plays twice in a row, rate limiting and circuit breaking.

Palantir Forward Deployed AI Engineer decomposition round: supplier-delay emails, LLM workflow MVP, and how to tell it is useful.

Cohere Integration/RL Research Engineer in three rounds: preference-graph consistency, offline preference batching, and a post-training eval harness.

Meta Business Support Engineer for Business Agent in three rounds: a tool-call parser, trajectory loop detection, and a tool-execution gateway design.

Netflix Open Connect Platform L5 in three rounds: a hash-ring lookup, a heartbeat liveness check, and a fleet desired-state control plane.

Microsoft Windows 365 / AVD Software Engineer II in three rounds: a stack calculator, best-fit Cloud PC host placement, and a pooled-desktop connection broker.

Meta Central Integrity infrastructure SWE in five rounds: weighted random pick, abuse-ring connected components, and an integrity decision system design.

Amazon SDE, AWS Agentic AI: recruiter screen and two technical interviews. Streaming tool call assembly, picking from hundreds of tools, a mostly-BQ round.

AMD AI Triton Kernels engineer: OA plus three interviews, with reshape inference, a broken Triton softmax kernel, and a memory ordering question.

Google Cloud Storage SWE III onsite in four rounds: equal-endpoint max subarray, folder-tree deletion order, C++ noexcept, and an object-store metadata design.

A Snowflake software engineer virtual interview in three rounds: a priority task executor, a multi-tenant job scheduler with leases, and PB-scale data sync.

An Optiver software engineer intern OA: a 90-minute market data processor with sequence buffering, snapshot recovery, per-symbol state, and Zap-N.

A Meta Systems ML Tooling technical interview with no coding: shared_ptr, virtual memory, RSS growth, profiling, locks, and AI-assisted leak hunting.

An Amazon AWS AI Infrastructure SDE virtual interview: vLLM PagedAttention, shipping model weights to thousands of nodes, rollout, and autoscaling.

A Datadog software engineer technical screen: one Python threading pub-sub question pushed through lock scope, slow subscribers, and a full queue.

A Goldman Sachs Bellevue Engineering New Analyst interview experience: HackerRank, then three Superday rounds on LRU concurrency, database locking, and Kafka.

A Google software engineer virtual onsite experience: a restaurant waiting list with FCFS deletes, then Meeting Rooms II with per-meeting room assignment.

Amazon SDE OA: 100-minute coding with branch-merge DP and an AI-assisted Movie DB repo fix, then Work Simulation and Work Style.

Anthropic General SWE, 2026 Q1: a web crawler coding screen covering BFS, concurrent deduplication, worker termination, and retry handling.

A short North America TikTok AI Agent and Automation loop that stopped after one technical round: harness, runaway execution, tool-registry drift, coupons.

A TikTok AI Agent software engineering intern technical screen covering agent evaluation, tool-call debugging, systems fundamentals, and directed-graph coding.

A Meta Systems ML interview recap covering inference scheduling, model-shard caching, ranking design, coding follow-ups, and performance trade-offs.

A five-round Harvey Agents software engineer interview covering document chunk merging, RAG, long-running research agents, project depth, and ownership.

A Sift Senior Backend Engineer interview experience covering telemetry, an investigation agent, concurrency, observability, and behavioral rounds.

A 45-minute Google L4 software engineer interview experience covering graph dependencies, cycle detection, communication, testing, and follow-ups.

A senior software engineer's Plaid AI Tooling interview experience: agent permissions, workflow state, retries, idempotency, system design, and coding.