Just finished the HF loop for a backend SWE role (not ML-specific, more infra/platform-adjacent). Writing this up because I couldn't find a solid end-to-end breakdown anywhere when I was prepping.
Total timeline was about 5 weeks from first recruiter contact to verbal offer.
Round 1: Recruiter screen (30 min) Standard intro stuff. They asked about my background, why HF, and did a quick gut-check on my interest level in open-source AI tooling. Nothing technical here.
Round 2: Technical phone screen (45 min) One coding question, medium difficulty on Leetcode's scale, graph-adjacent. They use CoderPad. Interviewer was engaged, asked follow-up questions about time complexity and what edge cases I was thinking about. Not a gotcha vibe at all.
Round 3: Home take-home (3-4 hours) This surprised me. They sent a small project, something like "extend this API endpoint and add tests." Realistic codebase, not a toy problem. They explicitly said to time-box it.
Round 4-6: Onsite (virtual, 3 sessions spread across 2 days) System design: 1 hour, I was asked to design a feature store or data pipeline component. Very ML-adjacent even for a backend role. Coding: 1 hour, two medium problems, one involved string manipulation the other was a BFS variant. Behavioral/values: 45 min with a senior IC and the hiring manager. More conversational than structured.
Feedback loop was fast, I heard back within 4 business days of the final round.
Overall: the process felt engineered by engineers who actually care. Less theater than my loops at bigger companies. The take-home was the unexpected part but honestly not terrible. If you're prepping, brush up on graph traversal, be ready to talk ML system concepts even if you're applying as a pure backend SWE, and have genuine opinions about open-source AI tooling.