Completed a 5-round loop for a senior MLE role about 6 weeks ago. Sharing because I couldn't find much specific info going in.
Round 1: 30 min recruiter screen. Standard background questions, they asked about my experience with NLP specifically vs. general ML. If you've only done vision or tabletop ML, that gap comes up fast.
Round 2: Take-home. They gave me access to the Cohere API and asked me to build something. Not a toy example, a real thing with multiple API calls. I had 3 days. They said "make it something you'd actually want to exist." I built a document reranking tool for internal wikis. It wasn't perfect but I narrated my trade-offs in the readme.
Round 3: Technical deep dive on the take-home + general ML systems. They went hard on: latency vs. accuracy tradeoffs, how embeddings work under the hood, what I'd change in the take-home. This was the round that weeded people out I think.
Rounds 4-5: Behavioral + cross-functional. Questions were about working with customers (enterprise use cases), navigating disagreement with research leads, and a weird one about how I'd explain embedding drift to a non-technical stakeholder.
I got an offer. The comp was good but not Google-tier. The team felt sharp and less politics-heavy than my last role. Worth interviewing if you have real NLP depth.