How Perplexity Interviews in 2026: Process, Questions & What They Score

1. 01

   Build a simplified todo list for AI agents: create tasks with increasing IDs, look them up, and move status forward only. Then add task dependencies, where blocked tasks become ready when their prerequisites succeed and fail when any prerequisite fails, directly or indirectly. Then render the list in an LLM-readable format.

   Tests a forward-only state machine and dependency graph propagation that must stay efficient on large inputs.

2. 02

   Implement Task Dependencies and Failure.

   Tests dependency graph traversal and cascading failure propagation, with attention to efficiency.

3. 03

   In-Memory File System: Strict mkdir, touch, ls, rm, rmdir and a Command Runner.

   Tests path parsing and tree or hashmap modeling with strict error handling for mkdir, touch, ls, rm and rmdir.

4. 04

   In-Memory File System: Add cd and Relative Path Support.

   Tests relative path resolution and current-directory state layered on an existing tree model.

5. 05

   Implement an in-memory (Unix) file system.

   Tests trie or hashmap path modeling, OOP class design and edge cases on paths for mkdir, ls and file content methods.

6. 06

   Implement a time-versioned (timestamped) key-value store with restore.

   Tests timestamp-versioned storage design and restoring an earlier state while keeping reads efficient.

7. 07

   Design a CreditTracker with expirations.

   Tests stateful class design with time-based expiry, balance accounting and efficient removal of expired credits.

8. 08

   Implement a stream filter that processes text with stop words.

   Tests streaming data structure design: filtering stop words in a token stream with deque, heap and hashmap.

9. 09

   Build a byte tokenizer.

   Tests byte-level text encoding and tokenizer implementation, a core component of LLM input pipelines.

10. 10

    Implement an LRU cache with TTL expiration.

    Tests doubly-linked list plus hashmap design with per-key expiry, and heap-based expiration management.

11. 11

    Implement beam search decoding.

    Tests beam search over log-probabilities with pruning and heap operations, core LLM decoding knowledge.

12. 12

    Implement a resilient LLM provider pool.

    Tests fault-tolerant design for LLM calls: provider failover, retries and health tracking.

Reported problems, listed so you know what to expect. Practice them in your own editor or on the platform you prefer; SupaCV's practice mode coaches how you talk through them.

1. 01

   Design a real-time search answer engine.

   Tests freshness versus quality tradeoffs across real-time crawling, event detection, caching and LLM integration.

2. 02

   Design an end-to-end RAG (retrieval-augmented generation) pipeline, including citations and hallucination mitigation.

   Tests end-to-end retrieval design: hybrid retrieval, reranking, citations and hallucination mitigation.

3. 03

   Design an LLM serving system at low latency.

   Tests low-latency inference serving: batching, KV-cache management, model sharding and streaming token delivery.

4. 04

   Architect a streaming response handler for LLM output.

   Tests streaming architecture for token-by-token LLM output, including delivery latency and failure handling.

5. 05

   Build a batching and inference optimization layer for embedding models.

   Tests request batching and throughput versus latency tradeoffs when serving embedding model inference.

6. 06

   Design a ranking pipeline for a search engine.

   Tests multi-stage ranking design for search: retrieval, scoring, reranking and latency budgets.

7. 07

   Design a trending queries ranking system.

   Tests streaming aggregation and time-decayed counting to rank trending queries at scale.

8. 08

   Design personalized discovery recommendations.

   Tests recommendation pipeline design: candidate generation, ranking and personalization signals.

9. 09

   Design a search system that adapts to constantly changing user interests.

   Tests handling interest drift in search relevance with fresh signals and ranking updates.

10. 10

    Design a large-scale web crawler.

    Tests crawl scheduling, politeness and rate limiting, duplicate detection and freshness policy at billions of pages.

11. 11

    Design a search autocomplete system.

    Tests prefix matching with trie-style indexes, trending query detection and P99 latency constraints.

12. 12

    Design a Long-Running Query Service.

    Tests asynchronous job design for slow queries: queues, status reporting, retries and result storage.

13. 13

    Diagnose an overloaded Kubernetes cluster.

    Tests systematic debugging of an overloaded cluster using metrics, resource limits, scheduling and scaling.

14. 14

    Design a personal finance aggregator.

    Tests integration design for third-party financial data: ingestion, sync, consistency and security.

15. 15

    Design a real-time knowledge graph platform.

    Tests graph data modeling, real-time ingestion and query design at scale.

16. 16

    When designing a chat application, how would you decide between using SQL and NoSQL databases?

    Tests data model and access pattern reasoning: consistency, scale and query needs for a chat workload.

17. 17

    What techniques can be used to optimize write performance from memory to disk?

    Tests storage write-path techniques such as buffering, batching, sequential writes and write-ahead logging.

Reported problems, listed so you know what to expect. Practice them in your own editor or on the platform you prefer; SupaCV's practice mode coaches how you talk through them.

1. 01

   What metrics would you track to evaluate the performance of your ML pipeline?

   Tests choosing offline and online evaluation metrics for an ML pipeline.

2. 02

   How would you determine the right criteria for when to forget outdated data, update the model, and evaluate performance?

   Tests model retraining and data freshness policy, including drift detection and evaluation criteria.

3. 03

   What testing methodologies are commonly used to ensure reliability and performance of AI models in production environments?

   Tests production ML reliability practices such as shadow testing, canaries, monitoring and regression suites.

4. 04

   How do you design systems around LLM unreliability?

   Tests fault-tolerant LLM system design: output validation, retries, fallbacks and evaluation.

Reported problems, listed so you know what to expect. Practice them in your own editor or on the platform you prefer; SupaCV's practice mode coaches how you talk through them.

1. 01

   Tell me about a feature you shipped where you had to make a hard tradeoff between latency and accuracy.

2. 02

   Where does Perplexity lose to traditional search, and how would you fix it?

3. 03

   Tell me about a time you identified a problem no one asked you to fix.

4. 04

   What interests you about AI search?

5. 05

   Why do you want to work at Perplexity?