Forward Deployed Engineer (FDE) Interview Questions

LLM & GenAI Interview Questions

Tokens, context windows, prompting vs RAG vs fine-tuning, hallucinations, evals, cost and latency: the conceptual core every AI-native FDE loop probes first.

• Walk me through what actually happens when an LLM generates a response.OpenAI · Anthropic · Google
• Walk me through the transformer architecture end-to-end, minus the heavy math.OpenAI · Anthropic · Google
• What is a token, and why should an FDE care about tokenization?OpenAI · Anthropic · Cohere
• What is a context window, and what are its practical limits in production?OpenAI · Anthropic · Google
• Explain temperature and top-p. When would you run temperature 0 in an enterprise workflow?OpenAI · Anthropic · Cohere
• What are embeddings, and how do they make semantic search work?Cohere · OpenAI · Glean
• Why is the dot product the similarity score in attention and embeddings, and when should you normalize to cosine?Cohere · OpenAI · Anthropic
• What's the difference between a base model and an instruct model, and why would a customer care?Anthropic · Mistral · Cohere
• Why do LLMs hallucinate?Anthropic · OpenAI · Scale
• You're building a classifier that flags risky contract clauses. Do you optimize for precision or recall?Scale · Microsoft · OpenAI
• A customer wants the model to 'know our docs.' Prompting, RAG, or fine-tuning: how do you choose?OpenAI · Anthropic · Cohere
• Your prompt plus retrieved documents exceed the context window. What are your options and tradeoffs?OpenAI · Anthropic · Glean
• Why does vector search retrieve 'related but wrong' results, and how do you fix it?Cohere · Glean · OpenAI
• Give me three concrete hallucination mitigations, and the cost of each.Anthropic · OpenAI · Scale
• When does fine-tuning beat few-shot prompting? How much data do you need, and how do you prevent regressions?OpenAI · Cohere · Mistral
• What is 'lost in the middle,' and how does it change how you assemble context for a RAG system?Anthropic · Glean · OpenAI
• What are the failure modes of tool/function calling, and how do you handle them in production?OpenAI · Anthropic · Sierra
• You need guaranteed JSON from the model, but it keeps breaking the schema. What do you do?OpenAI · Anthropic · Scale
• When is an agent the wrong answer? Argue against using agents.Anthropic · Sierra · OpenAI
• A customer says your LLM app is too slow. Give me five levers to reduce latency, and their tradeoffs.OpenAI · Microsoft · Google

RAG & AI Agent Interview Questions

Retrieval pipelines, chunking, reranking, tool-using agents, guardrails, multi-tenancy and eval harnesses: the modal FDE design round at OpenAI, Anthropic, Sierra, Glean and Scale.

• Walk me through a RAG pipeline end-to-end, and tell me where it usually breaksOpenAI · Scale · Cohere
• How would you chunk documents for retrieval, and how do you know your chunking is good?Cohere · Scale · Harvey
• Keyword search vs vector search: what does each actually buy you in a RAG system?Glean · OpenAI · Cohere
• When is an agent the wrong answer? Argue against building one.Anthropic · Sierra · OpenAI
• How would you evaluate a RAG system before letting a customer launch it?OpenAI · Anthropic · Scale
• How would you chunk contracts, Slack threads, and PDFs full of tables: same pipeline or different?Harvey · Glean · Scale
• Design a hybrid retrieval stack: BM25, vectors, and a reranker. What does each stage rescue?Glean · Cohere · OpenAI
• A customer's RAG pilot answers only 60% of questions correctly. Diagnose it.OpenAI · Scale · Anthropic
• Retrieval fails on the customer's internal jargon. Fine-tune embeddings, add a reranker, or rewrite queries?Glean · Cohere · Databricks
• Design a customer-support agent with order-status and returns tools. How do you keep it safe and useful?Sierra · Decagon · OpenAI
• A brand wants an agent that never gives financial advice, stays on-voice, and never mentions competitors. Design the guardrails.Sierra · Anthropic · Writer
• Your agent workflow takes 30 seconds. Design the streaming UX, including what happens when it fails mid-stream.Vercel · OpenAI · Sierra
• Build doc-QA for healthcare where a wrong answer is worse than no answer. How do you make it reliably say 'I don't know'?Anthropic · Harvey · OpenAI
• Every sentence the model writes must link to a source span. Design the citation system.Harvey · Glean · Anthropic
• Design text-to-SQL for executives, where a wrong-but-plausible query must never mislead anyoneDatabricks · Snowflake · OpenAI
• A customer wants to expose internal APIs to a model via MCP. Design it safely: auth, least privilege, versioning.Anthropic · OpenAI · Glean
• Run an LLM over 10 million records nightly, within budget. Design the batch pipeline.Databricks · Scale · Anthropic
• Ingest 40 enterprise sources (SharePoint, Confluence, Jira, drives) with incremental sync and deletion handlingGlean · Databricks · Microsoft
• Design the full eval harness for an LLM product: offline goldens, LLM-judge, online A/B, CI gatesOpenAI · Anthropic · Scale
• Design RAG over 10M documents with per-user permissions: a user must never see data they can't accessGlean · Microsoft · OpenAI

Coding & DSA Interview Questions for AI Engineers

Multi-part practical builds (in-memory DBs, rate limiters, parsers) plus the LeetCode-medium staples, calibrated to OpenAI/Anthropic practical screens and Meta/Palantir classics.

• Two Sum: return indices of the two numbers that add to a targetMeta · Scale · Palantir
• Group anagrams: cluster a list of strings into anagram groupsMeta · Glean · Scale
• Merge overlapping intervalsPalantir · Meta · OpenAI
• Insert a new interval into a sorted, non-overlapping interval listPalantir · Meta
• Top-K frequent elementsMeta · Glean · OpenAI
• Count subarrays whose sum equals KMeta · Scale · OpenAI
• Design a class that returns the moving average of the last N values in a streamOpenAI · Glean · Retool
• Build a wc-lite: count lines, words, and characters in text, with flags, factored for extensionAnthropic · Retool · OpenAI
• Three Sum: find all unique triplets that sum to zeroMeta · xAI · Scale
• Design and implement an LRU cache with O(1) get and putPalantir · Meta · xAI
• Merge K sorted lists into one sorted outputMeta · Glean · Scale
• Number of islands: count connected regions in a gridMeta · Palantir · xAI
• Shortest path in a grid with obstaclesMeta · xAI · Palantir
• Word search: does a word exist as a path of adjacent cells in a letter grid?Meta · Palantir · xAI
• Serialize and deserialize a binary treeMeta · Glean · OpenAI
• Flight segments (start, end, seats): find the maximum simultaneous passengersPalantir · Meta
• Parse server logs and report the top-K users by error count per hour, handling malformed linesPalantir · Anthropic · Retool
• Sessionize a support-event stream: a gap over 30 minutes starts a new session; dedupe eventsGlean · Scale · Retool
• Build a CSV diff tool: report added, removed, and changed rows between two files, with composite keysPalantir · Retool · Scale
• Build a versioned key-value store: put/get, plus get(key, timestamp) for historical readsOpenAI · Glean · Scale

Machine Learning & Data Science Interview Questions

Classical ML, embeddings, metrics (precision/recall, AUC), experiment design and statistics: what Google Cloud, Databricks and Scale still test alongside GenAI.

• Explain precision, recall and F1, and which one matters for a fraud-detection customer?Google · Scale · Microsoft
• Your model has 0.95 ROC AUC. The customer is unimpressed in production. What does AUC actually tell you?Google · Scale · Meta
• Explain the bias-variance tradeoff, and how it shows up in a real customer deployment.Google · Microsoft · Databricks
• How do you detect overfitting, and walk me through the tools you'd use to fix it.Google · Databricks · Microsoft
• The customer's dataset is 99.5% negatives. Their model shows 99.5% accuracy. Walk me through what you'd do.Scale · Google · Microsoft
• Your model scored 95% in the pilot and 70% in production. What happened?Google · Databricks · Scale
• Explain k-fold cross-validation, and when would you refuse to use it?Google · Databricks · Microsoft
• A customer says 'we want the model to be accurate.' How do you turn that into the metric you'll optimize?Google · Databricks · Snowflake
• Explain gradient descent to me like I'm a smart engineer who's never done ML. What can go wrong?Google · Meta · Scale
• Tree ensembles vs linear models vs neural networks, how do you choose for a customer's tabular problem?Databricks · Google · Snowflake
• The customer wants to try a fancier model. You think the win is in feature engineering. Make your case, with examples.Databricks · Snowflake · Google
• What does cosine similarity actually measure, and why is it the default for comparing embeddings?Google · Databricks · Scale
• A customer wants to segment 2 million users. Walk me through clustering, k-means, choosing k, and where it goes wrong.Google · Snowflake · Microsoft
• When does classical ML beat an LLM, and when is the LLM the right call? A customer wants 'AI' for everything.Google · Databricks · Scale
• A regulated customer asks: 'Why did the model deny this application?' Explain SHAP and how you'd deploy explainability.Google · Microsoft · Databricks
• A customer wants to forecast weekly demand. What's different about time-series ML, and how do you avoid embarrassing yourself?Databricks · Google · Snowflake
• Your model says '90% probability.' The customer asks if that means it's right 90% of the time. Is it?Google · Scale · Microsoft
• Explain what a p-value actually is, to me, and then to a customer's VP of Marketing.Meta · Google · Microsoft
• Design an A/B test for a customer's new recommendation model. How long do you run it, and on how many users?Meta · Google · Databricks
• The customer checked the A/B dashboard daily and stopped the test the day it hit significance. What's wrong, and what do you tell them?Meta · Google · Snowflake

SQL & Data Engineering Interview Questions

Window functions, gaps-and-islands, Spark internals, lakehouse design, idempotent pipelines and CDC: the Databricks/Snowflake/Palantir data track.

• What's the difference between RANK, DENSE_RANK and ROW_NUMBER, and when does the choice actually matter?Snowflake · Meta · Retool
• Find the second-highest (or Nth-highest) salary per department.Meta · Microsoft · Retool
• Write SQL for the top 3 products by revenue in each region, per month.Snowflake · Databricks · Palantir
• A table has duplicate rows for the same business key. Write SQL to keep only the latest version of each.Retool · Meta · Palantir
• Why did my LEFT JOIN start behaving like an INNER JOIN after I added a WHERE filter?Retool · Meta · Microsoft
• Find users who logged in on 3 or more consecutive days (gaps-and-islands).Meta · Palantir · Databricks
• Sessionize a raw event stream in SQL: a gap of more than 30 minutes starts a new session.Databricks · Meta · Snowflake
• Write SQL for a signup → activation → purchase funnel, broken down by weekly signup cohort.Meta · Snowflake · Retool
• Build a cohort retention matrix in SQL: % of each monthly signup cohort still active N months later.Meta · Databricks · Snowflake
• After adding a join, your row count and revenue totals exploded. What happened and how do you detect it?Palantir · Retool · Meta
• Pivot rows to columns in SQL, without using the PIVOT keyword.Snowflake · Retool · Microsoft
• You're handed raw JSON events in a Snowflake VARIANT column. How do you query and model them?Snowflake · Databricks · Palantir
• A customer's query on a 2-billion-row table takes 40 minutes. Walk me through your diagnosis.Databricks · Snowflake · Palantir
• Explain Snowflake micro-partitions and clustering, and when a bigger warehouse will NOT make a query faster.Snowflake · Microsoft · Palantir
• Why is Parquet the default for analytics? What does columnar storage actually buy you over CSV?Databricks · Snowflake · Palantir
• How do you size Snowflake warehouses, when do you scale up vs scale out, and how do you keep the bill sane?Snowflake · Microsoft · Retool
• Spark: what's the difference between wide and narrow transformations, and why are shuffles expensive?Databricks · Palantir · Microsoft
• Your Spark join is skewed, 199 tasks finish in seconds, one runs for an hour. Fix it.Databricks · Palantir · Microsoft
• A customer's Delta table has 4 million small files and every query crawls. What happened, and how do you fix it?Databricks · Palantir · Snowflake
• Design a star schema for a ride-sharing company's analytics.Databricks · Snowflake · Palantir

System Design Interview Questions for AI/ML

Palantir-style decomposition, productionizing POCs, debugging customer incidents, observability, VPC/air-gapped deploys: where demos become deployments.

• What's the difference between at-least-once and exactly-once delivery, and why should an FDE care?OpenAI · Retool · Databricks
• Explain how rate limiting works, fixed window, sliding window, token bucket. When does each break?OpenAI · Anthropic · Glean
• What do p50, p95, and p99 latency actually mean, and why do production engineers obsess over the tail?Vercel · OpenAI · Microsoft
• What is a dead-letter queue, and how do you handle a poison message in production?Retool · Microsoft · Databricks
• It's Friday evening at a customer site and they want a hotfix shipped now. How do you do it safely, and when do you refuse?Palantir · Retool · Microsoft
• Estimate the capacity and cost of an app with 50k DAU making 10 LLM calls each. What do you provision for?OpenAI · Anthropic · Microsoft
• It's 9am Monday and p99 latency is 10x normal. Walk me through your first 30 minutes.OpenAI · Vercel · Microsoft
• A customer's app calling your API times out intermittently. You can't see their code. Debug it.OpenAI · Anthropic · Retool
• A Python service starts double-processing messages under load. Why does this happen, and how do you fix it?OpenAI · Retool · Databricks
• A customer's Next.js site has a 3-second TTFB. Diagnose it and walk me through fixes, and how you'd prove each one worked.Vercel · Retool
• Design observability for an LLM application. What do you log, trace, and alert on, and how is it different from normal services?OpenAI · Anthropic · Glean
• How do you version and roll out prompt changes like code, review, canary, rollback, audit?Anthropic · OpenAI · Glean
• Design a job queue for long-running LLM tasks, priorities, cancellation, progress, and poison messages.OpenAI · Anthropic · Retool
• Your product deploys into dozens of customer environments. Design secrets management across all of them.Palantir · Retool · Microsoft
• You get a vague prompt like 'design a system to detect fraud.' What do you do in the first ten minutes?Palantir · OpenAI · Microsoft
• Decompose: design a system to detect credit-card fraud.Palantir · Databricks · Microsoft
• Decompose: design a system to coordinate disaster response for a city.Palantir · Microsoft
• Decompose: merge patient records across a network of hospitals.Palantir · Databricks · Microsoft
• Decompose: design a system to route a fleet of delivery trucks.Palantir · Microsoft · Databricks
• Decompose: predict flight delays, and survive the interviewer changing the rules every five minutes.Palantir · OpenAI · Databricks

Behavioral & Customer Interview Questions for FDEs

Discovery role-plays, hostile-stakeholder scenarios, demo recovery, panel presentations and the 'why customer-facing' filter: the highest-variance, least-prepped FDE rounds.

• Tell me about the most ambiguous project you've owned end-to-end. What did you do in week one?Palantir · OpenAI · Scale AI
• Why customer-facing instead of pure engineering? Convince me you won't quit in a year.Palantir · OpenAI · Databricks
• This role is 50–75% travel, sometimes living at customer sites for weeks. Honestly, how will you handle that?Palantir · Databricks · Snowflake
• Tell me about a time you shipped in days something that 'should' have taken months. Which corners did you cut, and how did you choose?OpenAI · Scale AI · xAI
• Tell me about learning an unfamiliar domain fast. How fast, and how did you validate you understood it?Harvey · Palantir · Scale AI
• Describe your worst production incident at a customer. What did you tell them, and when?Palantir · OpenAI · Microsoft
• Tell me about a time you said 'no' to a customer, and they thanked you for it later.Databricks · Snowflake · Sierra
• Tell me about a project that failed. Whose fault was it?Palantir · OpenAI · Google
• Estimate: how many LLM tokens per day would a Fortune-500 customer-support org consume?Google · Palantir · Microsoft
• Tell me about turning a skeptical stakeholder into a champion.Databricks · Microsoft · Palantir
• Role-play: I'm a VP at a regional bank and I want 'a chatbot.' Run the discovery call.Sierra · OpenAI · Anthropic
• Explain RAG to our Chief Legal Officer in two minutes. She's brilliant, busy, and non-technical.Harvey · Anthropic · OpenAI
• Role-play: a CTO tells you 'we tried GPT last year, it hallucinated all over our data, AI doesn't work.' Respond.OpenAI · Anthropic · Scale AI
• An exec asks point-blank: 'Can it be 100% accurate?' Answer without lying, and without losing the deal.Harvey · Anthropic · Sierra
• Your customer says the pilot 'should just be good.' Define success metrics with them, live.OpenAI · Scale AI · Databricks
• A customer arrives with 30 AI use cases. Design the workshop that picks the first one.Palantir · Microsoft · Scale AI
• Here's an engineer's status update full of jargon. Rewrite it as the steering-committee email.Palantir · Microsoft · Databricks
• Your demo breaks in front of the customer, mid-demo. What do you do in the room?Snowflake · Vercel · Google
• The customer insists on fine-tuning when RAG clearly fits. They won't budge. Trusted advisor or vendor, what do you do?OpenAI · Cohere · Databricks
• You're alone on-site. The customer asks for something out of scope, and your team is asleep in another timezone. Decide now.Palantir · Scale AI · OpenAI

MLOps & ML Engineering Interview Questions

CI/CD for models, drift detection and retraining, Kubernetes inference, feature stores, staging-to-production promotion and pipeline testing: what AWS, Databricks and every ML-platform loop drills.

• How is CI/CD for ML models different from traditional DevOps CI/CD?Amazon · Microsoft · Capital One
• What are your day-to-day responsibilities as an MLOps engineer?Amazon · Microsoft · JPMorgan
• What is MLflow for, and what are its four components?Databricks · Amazon · Microsoft
• Why use a model registry instead of just storing model artifacts in S3?Amazon · JPMorgan · Capital One
• What problem does a feature store solve? Explain the online store versus the offline store.Uber · Capital One · Netflix
• What's the difference between shadow deployment and A/B testing a model?Netflix · Capital One · Google
• When you say you 'push staging to production,' what does that actually mean for a model?Amazon · Databricks · Microsoft
• How did you ensure quality and testing on top of your MLOps pipeline?Amazon · Google · Databricks
• How do you ensure reproducibility in ML workflows?Amazon · JPMorgan · Databricks
• Walk me through promoting a model from staging to production with the MLflow Model Registry.Databricks · JPMorgan · Capital One
• How would you implement an MLOps pipeline on AWS using SageMaker, CodePipeline, and Lambda?Amazon · Capital One · JPMorgan
• Explain MLOps on Azure, how do Azure DevOps and Azure ML Pipelines fit together?Microsoft · JPMorgan
• Your pipeline retrains automatically. How do you decide whether the new model replaces the old one?Amazon · Uber · Capital One
• Design an automated retraining pipeline. What should trigger retraining?Uber · Capital One · Amazon
• What is point-in-time correctness, and how do you avoid leakage in continuous retraining?Capital One · JPMorgan · Uber
• What is training-serving skew, and how do you keep online and offline features consistent?Google · Uber · Microsoft
• Design a feature store for a real-time fraud detection model.JPMorgan · Capital One · Uber
• Which statistical tests would you use to detect drift, and what thresholds should trigger action?Microsoft · JPMorgan · Amazon
• How do you monitor a model in production when ground-truth labels arrive late, or never?Netflix · Uber · JPMorgan
• Would you choose blue-green or canary deployment for a new model version, and why?Capital One · Amazon · Netflix

ML Infrastructure & GPU Interview Questions

GPU/TPU workloads, distributed training and parallelism, inference serving (vLLM, batching, KV cache), cluster scheduling and scaling API gateways: the infra depth NVIDIA, Google and the AI labs probe.

• Explain how the CUDA execution model maps to hardware, grids, blocks, warps, SMs.NVIDIA · Google · CoreWeave
• Walk me through the GPU memory hierarchy, registers, shared memory, L2, HBM. What lives where and why?NVIDIA · xAI · Meta
• What is warp divergence and why does it hurt performance?NVIDIA · xAI
• Explain memory coalescing and shared-memory bank conflicts. How would you fix a kernel that has both?NVIDIA · Meta · xAI
• What is occupancy, and how do you balance it against register and shared-memory usage when choosing block size?NVIDIA · Google · CoreWeave
• How do you determine whether a kernel is memory-bound or compute-bound?NVIDIA · Google · Together AI
• Have you worked on machine learning GPU workloads?NVIDIA · CoreWeave · Together AI
• Compare data, tensor, and pipeline parallelism, when do you use each, and how do they combine into 3D parallelism?OpenAI · Anthropic · Meta
• Explain how ring all-reduce works and derive its communication cost.NVIDIA · OpenAI · Meta
• What do ZeRO and FSDP actually shard, and how much memory does each stage save? Where does gradient checkpointing fit?Meta · OpenAI · Anthropic
• We need to train a 100B-parameter model that won't fit in memory. Design the data and model parallelism.Google · Anthropic · OpenAI
• Pipeline parallelism, does it run over the scale-out network or stay inside the node? Map TP, PP, and DP onto the fabric.xAI · OpenAI · NVIDIA
• What does NCCL actually do, and why can GPU utilization read 100% while the job is communication-bound?NVIDIA · CoreWeave · xAI
• InfiniBand vs RoCE for a GPU training cluster, how do you choose?xAI · Meta · CoreWeave
• How do you handle expert load balancing in an MoE model? What happens if 90% of tokens route to 10% of experts?OpenAI · xAI · Together AI
• Your training run compiles and runs, but the loss is flat. Walk me through your debugging process.Google · Anthropic · OpenAI
• How do you keep a multi-week training run alive across hardware failures and stragglers?Anthropic · CoreWeave · xAI
• What experience do you have with ML infrastructure, including distributed GPU clusters?CoreWeave · xAI · OpenAI
• What is the KV cache, why does it dominate serving memory, and how do you size it? Do the math for a 70B model.OpenAI · Anthropic · NVIDIA
• How does vLLM's PagedAttention work, and what problem does it actually solve?Together AI · NVIDIA · Anthropic

AI Security & Prompt Engineering Interview Questions

Prompt-injection defense in depth, PII handling, consumption tracking and abuse prevention, audit trails, SOC2/EU AI Act and enterprise data-governance questions every CISO-facing FDE must survive.

• Explain direct vs indirect prompt injection, and how each one actually breaks an LLM application.OpenAI · Microsoft · Glean
• How is a jailbreak different from prompt injection? Give an example of each from a real application.Anthropic · Microsoft · Scale
• Why do system prompts leak, and what should never be in one?OpenAI · Anthropic · Salesforce
• Walk me through the OWASP LLM Top 10, which two risks would you prioritize for an enterprise agent deployment?Microsoft · Palantir · Scale
• What is 'excessive agency' in an agentic system, and what controls actually limit it?OpenAI · Anthropic · Salesforce
• What is data/model poisoning, and how would you detect a backdoor introduced through fine-tuning or embedding data?Microsoft · Scale · OpenAI
• Explain MITRE ATLAS. How would you map an observed attack on an AI system to its tactics and techniques?Microsoft · Palantir · Scale
• Your LLM agent is vulnerable to prompt injection that reveals the system prompt. How do you defend it?OpenAI · Anthropic · Salesforce
• How would you defend a RAG system against poisoned retrieved context and vector/embedding attacks?Glean · Microsoft · Anthropic
• How do tool poisoning and prompt injection apply to MCP servers and agentic AI, and how do you defend them?Anthropic · OpenAI · Salesforce
• Walk me through how you would red-team a customer-facing GenAI chatbot.Microsoft · Scale · OpenAI
• Design an ongoing AI red-team program: team, harm categories, cadence, and what you'd automate with PyRIT first.Microsoft · Scale · Anthropic
• Design safety guardrails into the request flow of an LLM serving system, where do they sit and what does each catch?Anthropic · OpenAI · Salesforce
• Compare PII detection methods, regex, entropy, NER, external APIs. Why can't regex alone find a person's name?Microsoft · Snowflake · Glean
• Should PII redaction live at the AI gateway or in the application layer? Argue the trade-offs.Microsoft · Salesforce · Glean
• Design a PII redaction system for an LLM pipeline, PII enters at user input, retrieved context, tool results, and output. Guard all four.Palantir · Microsoft · Glean
• Prompts and outputs contain personal data. What does GDPR mean for each call to an external LLM endpoint?OpenAI · Anthropic · Microsoft
• The customer asks: 'Will you train on our data?' Give the precise answer, and explain zero-data-retention.OpenAI · Anthropic · Glean
• The customer says 'our data can't leave our environment.' Walk me through the deployment options and what each really buys.Anthropic · Palantir · Microsoft
• A hospital wants an LLM feature over patient records. What does HIPAA actually require of your design?Microsoft · Palantir · Salesforce

ML System Design Interview Questions

Design the model behind a product surface: feed ranking, recommendations, watch-time and ETA prediction, fraud and bot detection, ads ranking. Candidate generation, feature and label design, training, offline and online evaluation, and serving at scale, the way Meta, Google, Netflix and recsys teams run the round.

• Design the ranking model for a personalized feed (Instagram-style).Meta · LinkedIn · Pinterest
• Design a music recommendation system (Spotify-style).Spotify · Apple · Amazon
• Design an evaluation framework for an ads-ranking system.Meta · Google · Amazon
• Predict watch time for items in a video catalog, Netflix-style. How do you build it?Netflix · YouTube · Disney+
• Design a system to detect bots and inauthentic accounts in real time.Meta · Roblox · Google
• Design a CTR and conversion-rate prediction system for ads.Google · Meta · Amazon
• Design an ETA prediction system for a maps or navigation app.Google · Uber · DoorDash
• Design a landmark or image recognition system at scale.Google · Apple · Pinterest
• Build a fraud-detection model for payments.Stripe · PayPal · Adyen
• Design a recommendation engine for an online-course (or e-commerce) catalog.Amazon · Coursera · Udemy
• Design the 'For You' ranking system for a short-video feed.TikTok · Meta · YouTube
• Design learning-to-rank for product/marketplace search.Amazon · DoorDash · Etsy

FDE interview questions by company

Each company runs a different loop. These pages break down the process, rounds, and what each interviewer is really testing, compiled from public reports.

FDE interview questions: FAQ