Let me be direct: the LinkedIn MLE loop is heavily shaped by what team you're targeting. I went through it for a role on the feed ranking team, which is recsys. If you're interviewing for a different ML area (NLP, CV, trust/safety) some of this will still apply but the system design prompt will obviously differ.
Five rounds after the initial technical phone screen.
Technical phone screen: One coding problem (medium, sorting + hashing), one ML question. The ML question was: you have a binary classifier for whether a user will click on a job posting. Your precision is high but recall is low. Walk me through why that might be and how you'd investigate. Classic imbalance question. Make sure you can talk about threshold tuning, class weights, and data-level approaches without getting confused between them.
Onsite:
Round 1, coding: two problems, both medium. One was a sliding window, one was a trie problem. Standard algorithm prep is sufficient. They don't seem to care that much about graph problems for this team.
Round 2, ML fundamentals: pure theory round. Topics covered: bias-variance tradeoff applied to a real scenario, how gradient boosting works and when you'd prefer it over a neural net, L1 vs L2 regularization effects on sparse features (very relevant for their ad models), and how you evaluate a ranking model when you can't A/B test (offline metrics: NDCG, MAP, MRR). You need real depth here, not interview-prep-surface knowledge.
Round 3, ML system design: design the feed ranking system for LinkedIn. This is their home turf so they know the problem cold. Start with user objectives vs engagement objectives tension. Talk about candidate generation vs ranking stages. Feature engineering: what signals matter (network graph features, content features, freshness decay). They pushed on: how do you handle cold start for a new post? How do you prevent engagement optimization from rewarding clickbait? I got into diversity injection and topic coverage constraints. They seemed to like specifics.
Round 4, behavioral: standard STAR. They asked about disagreeing with a research direction and how you navigated it. And about a project that didn't work out and what you learned.
Round 5, cross-functional: a round with a data scientist and a PM from the team. They're checking that you can translate between research and product requirements.
Total process: 6 weeks. Offer came in at E5. Base was around $215k in Sunnyvale, RSUs on top.