Interview questions
Meta AI engineer interview questions (2026)
10 real interview questions reported by engineers who interviewed at Meta, spanning ML Engineering, AI Engineering, Data Engineering, Software Engineering. Every question is scored against a golden answer on the things Meta actually grades, architecture, token efficiency, security and correctness, not just whether your code runs.
Meta Software Engineering questions
- LLM output validation with schema enforcement and type checking
Implement a top-level `solve(payload)` that validates an LLM's JSON output against a simple schema. ## The solve contract `payload` is a dict: `{"raw": str, "schema": {field: typ
- LLM Cache Bug: Same Response Returned Regardless Model Parameter
A production LLM response caching layer is returning wrong results for some users. When two engineers call the same prompt with different models, they get the same response back re
- Async batch processor silent failures with missing error handling logic
Your team built an async batch processor that sends multiple prompts to an LLM in parallel. In production, some results are None with no error logged. Find the bug.
- LLM retry logic causes infinite loops on malformed API responses
An LLM customer support agent works perfectly on normal queries but certain edge case inputs are causing runaway API calls and billing spikes. The on-call engineer got paged at 3am
Meta ML Engineering questions
- Distributed Training — Fitting and Scaling a 70B Model
Your team must train a 70B-parameter model; a single H100 has 80GB. Show why the model plus optimizer state does not fit on one GPU (FP16 weights, gradients, Adam state); compare d
- Distributed Training — ZeRO Stages and Communication Tradeoffs
You are choosing a memory-optimization strategy. Explain ZeRO stages 1, 2, and 3 (what each shards and the memory vs communication tradeoff); when you would prefer tensor paralleli
- Medical LLM Governance: Audit Infrastructure at Meta Scale
Design AI governance and audit infrastructure for a medical LLM at Meta scale Meta is deploying a large language model for medical information to 500M users. Regulators require: -
Meta AI Engineering questions
- Fine-tuned model tokenizer mismatch with deployment environment configuration
After fine-tuning a Llama model on domain-specific data, the deployed model produces garbled outputs that look like random tokens. The fine-tuning metrics looked great. Find the de
- Next Word Prediction System Architecture Design
Design a next word prediction system.
Meta Data Engineering questions
- Feature drift detection pipeline with automated monitoring and alerting system
A data scientist reports that a feature in the feature store has drifted. Walk through the pipeline you would build to detect and alert on this automatically.
Prepare for your Meta interview
Velocode turns reported Meta interview questions into scored practice. Free accounts get the full community problem set and one Meta-tagged question per domain; Pro unlocks every company-verified question, the interview simulator, and your domain readiness radar.