LinkedIn · Primly Community

LinkedIn onsite / final round, how it really goes in 2026 (virtual format)

infra_ines · 5 replies

Did the LinkedIn virtual onsite for an L5 backend role in March 2026. Four rounds across two half-days. Here's the structure and what I'd do differently.

Day 1: Round 1 (coding, 45 min): Two medium problems. First was dynamic programming, second was a graph BFS problem. Interviewer asked me to explain time/space complexity for each after I coded it. Not hard but you can't skip the analysis. Round 2 (system design, 60 min): I got "design a job recommendation system for LinkedIn." Good prompt for the company, forces you to think about their actual product. I covered data ingestion, candidate-job matching, personalization, and serving infrastructure. They specifically asked how to handle cold start for new users.

Day 2: Round 3 (behavioral, 45 min): Two interviewers, took turns asking questions. I got conflict resolution, a prioritization question, and one about giving feedback to a peer. Felt very rubric-based. Round 4 (coding, 45 min): One harder problem, a bit of a tree + DP hybrid. More like a Leetcode hard in structure but not impossible. I didn't fully solve it, got most of the way there, and the interviewer asked me to walk through what I'd do next.

What I learned: Not solving a problem 100% in the onsite is not automatically disqualifying. I think they care about your process, communication, and whether you identify the right approach even if the code isn't perfect. I partially solved round 4 and still got the offer.

Debrief timeline: Recruiter called 5 business days after day 2. No interim updates, which was nerve-wracking. Verbal offer first, then written within 24 hours. Negotiated base up about $12K from the initial number.

Total onsite experience: 2 days, 3.5 hours of interviews. Virtual made it less tiring than an in-person trip but it's still a long haul.

5 replies

infra_ines

The job recommendation system prompt is so on-brand. It makes prep easier honestly, you can actually use LinkedIn to think through edge cases. What did you say for cold start? That's the interesting part.

remote_swe_42

I covered a few options: use onboarding signals (role, industry, location they fill in), content-based filtering from job title/description matching, and falling back to popularity-based recs by location. The interviewer then asked how I'd bootstrap the ML model for new users which led into a decent conversation about transfer learning from similar users.

staff_steph

Five business days for debrief is actually normal but it feels like a year. Did they give you any signal during the call about which way it was going, or was it just "we'll be in touch"?

frontend_fran

Helpful to know partial solutions don't kill you. I got stuck in my last onsite and almost gave up instead of narrating my thinking. Would have been better to talk through my reasoning even without finishing.

content_cole

How hard was it to negotiate the base? Did they come back quickly or did it take multiple rounds?