NLP Engineer Big Tech Insider
$19.99
Product preview

NLP Engineer Big Tech Insider

$19.99

NLP interviews at big tech are now LLM interviews


In 2026 loops, an NLP Engineer at Google is asked to implement multi-head attention and an LM head live. At OpenAI it is debugging a transformer with four planted bugs, then adding a KV cache. At Anthropic it is a longest-match tokenizer. The classic questions about SVMs and PCA are mostly gone.

We spent weeks collecting what candidates actually report from NLP and LLM loops at 11 companies between 2022 and 2026. The result is NLP Engineer Big Tech Insider, a 117-page guide built for 2027 loops. This post explains what is inside and previews 50 of its 130 questions.

Independent guide. Not affiliated with or endorsed by any company named.


What's inside the guide

Every question is a card with a senior-level answer, and every card shows where the question came from and when.

• 130 questions across 11 companies: Google, Meta, Amazon, Apple, Microsoft, NVIDIA, OpenAI, Anthropic, Bloomberg, TikTok and Netflix, plus Databricks, Uber, LinkedIn, Spotify, Grammarly and Duolingo.

• Honest source tags: reported by candidates, from interview-prep vendors, drawn from public job postings, or our own practice questions. Each card carries its source and year.

• Structured answers: a short answer, the reasoning, and capacity numbers on design questions (QPS, latency, memory).

• Next probes: three follow-ups per card, the questions the interviewer asks after your first answer.

• Tested code: every coding answer runs against a test suite, with KV cache, BPE, layer norm and sampling checked against reference implementations.

• 14-day plans: one per company, each with a must-do five.

• 20 live NLP jobs: taken from company careers pages and checked on 12 Nov 2026, with posted pay ranges.

• Roadmap and tech stack: what to learn, in order, and the tools these teams list in postings.

• Behavioural templates: Amazon Leadership Principles and values rounds, written as templates to fill with your own stories, not scripts.

How loops differ by company

The rounds look similar on paper, but each company weights them differently. The costliest prep mistake is getting the format wrong: skipping ML coding for Google, Leadership Principles for Amazon, or implementation depth for OpenAI.

• Google: two ML coding rounds focused on NLP, one LeetCode, one NLP system design and one leadership round. Expect to implement MHA, an LM head or a tokenizer, and to design a QA system.

• Meta: coding, behavioural and ML system design. Harmful content detection and LLM assistant or agent design come up most.

• Amazon: coding, ML breadth, ML depth and Leadership Principles in STAR format. Transformers, BERT vs RoBERTa and seq2seq, with LP stories in every round.

• Apple: coding, NLP and ML, with four to five sessions for NLP Engineer roles. LSTMs, localisation, RAG and agents, and on-device inference.

• Microsoft: NLP concepts, a LeetCode medium and an applied NLP problem. Summarisation, hallucination, catastrophic forgetting and metrics.

• OpenAI: ML fundamentals, transformer debugging and distributed coding. Debug a transformer, build a KV cache, design an LLM API rate limiter.

• Anthropic: online assessment, technical screen and a values round. Tokenizers, BPE, KV cache, LLM serving and prompt caching.

• Bloomberg: phone rounds, a 30 to 60 minute research talk and an onsite. Spam filters, conversation threading and vanishing gradients.

• NVIDIA: LeetCode, a transformer deep dive and project deep dives. Tokenization, parallelism, quantisation and inference cost.

• TikTok: project deep dive, LLM fundamentals and LeetCode, often dynamic programming. Vocabulary space, SFT vs LoRA, LSTM vs GRU.

• Netflix: general ML plus a LeetCode medium. Knowledge graphs for entity resolution.

Teams inside one company run different loops, so confirm the format with your recruiter.

Meet the creator