Just accepted an offer last month. I was in prep mode for about six weeks. Here's the honest version of what worked and what I wasted time on.
What I'd do first: Understand their actual product and business model before touching a single LeetCode problem. The Palantir pitch round can tank your onsite if you go in cold. Spend 4-5 hours on their investor materials, customer case studies, and any YouTube talks by their engineering team. This isn't about memorizing facts. It's about being able to speak fluently about enterprise data challenges, which is the air they breathe.
Coding prep. I used LeetCode, targeting mediums in graphs, trees, and dynamic programming. The thing that actually helped was recording myself solving problems and watching it back. You notice when you go silent, when you stop narrating your thought process. Palantir interviewers want to follow your reasoning. If you get stuck and say nothing for 90 seconds that's a bad sign even if you eventually solve it.
System design. I did 8 mock sessions over 4 weeks. Focus on scoping. They're not impressed by complexity, they want to see you make deliberate trade-offs and explain them. Designing a URL shortener? Fine. But be ready to defend every choice and cut scope when asked.
Behavioral. Four or five strong STAR stories will cover most of what they ask. Conflict with a coworker, project you'd do differently, time you took initiative, time you had to influence without authority. Have numbers in your stories. "Reduced latency by 40%" beats "improved performance significantly."
What I wasted time on: LeetCode hards. I did a bunch and they never came up. Mediums with clean implementations were what actually showed up in my rounds.
The loop is thorough but it's learnable. Six weeks of focused prep is enough if you're targeting the right things.