ok so i'm three weeks out from getting an offer from HF for a new grad SWE role and i feel like i owe this thread to the internet because i was completely lost going in.
first thing: the new grad pipeline at HF is smaller and less standardized than big tech. they don't have a formal university recruiting program the way Google or Meta does, or at least not in 2026. i applied through the website, got a reply after about 3 weeks (yes it's slow), and the loop was 3 rounds.
what they actually tested:
algorithms/data structures: yes, they do leetcode-style problems, but they skew medium. i got a graph traversal problem (BFS variant) and a string manipulation problem. no dynamic programming deep end. they want to see clean code and that you can think through edge cases out loud. python is totally fine.
system design (light version for new grads): don't panic. they're not expecting you to design distributed caching systems. i got "design a simple rate-limiter for an API" and we spent about 20 minutes on it. they just want to see that you can reason about real-world constraints and ask good clarifying questions.
behavioral: one round with a hiring manager. they asked about a project i built (be ready to go really deep on one or two projects from your portfolio or coursework), a time i disagreed with a technical decision, and why HF specifically. the last one matters. you need to have a genuine answer about the open-source ML ecosystem, not just "i like transformers."
prep tips that actually helped: grind mediums on neetcode, not random leetcode. patterns > volume. read the HF blog and actually use the library. if you've fine-tuned something with transformers or used datasets/evaluate, talk about it in detail. have an opinion about something in the ML ecosystem. they're building tools for researchers; they want people who engage with that world.
timeline was slow (applied in jan, offer in late march) but the process itself once it started moved in about 2.5 weeks. don't give up if you're waiting on the first response.