Cohere · Primly Community

Cohere new grad / entry level interview: how to prep, what I wish I'd known

jp_newgrad · 7 replies

I went through the Cohere new grad / entry level interview process this spring as a 2025 grad and I want to write down everything I had to piece together from scattered sources.

First: Cohere doesn't post a ton of dedicated new grad roles but they do have junior SWE and junior research positions that are realistically entry-level. The recruiter was upfront that the bar is the same as senior-adjacent roles in terms of coding quality, just scaled in complexity. That was a bit alarming to hear as a new grad but honestly it was true.

What the loop looked like for me: Online assessment. Two LeetCode-style problems, 75 minutes. I got a medium graph problem and a medium-hard dynamic programming problem. This was the hardest filter and where I think most new grads get cut. Technical phone screen with a senior engineer. More coding, plus conceptual questions about ML fundamentals. They asked me to explain transformers at a high level and how attention works. Not expecting a PhD answer, just want to know you're not allergic to the domain. Final virtual onsite: three rounds. Coding (similar difficulty, one system design lite), a research discussion where you walk through a project or paper you know well, and a behavioral round.

How I prepped:

Algorithms: Neetcode 150, leaning hard on graphs, DP, and sliding window. The online assessment is the real gate so treat it like the most important round.

ML fundamentals: read the original Attention is All You Need paper at a surface level, watched some 3Blue1Brown videos, made sure I could explain embedding spaces, tokenization, and inference at a conceptual level. Didn't need to implement anything from scratch.

Behavioral: they asked about a time I had to communicate a complex technical concept to a non-technical audience. classic but I wasn't ready for it in round 3.

One thing nobody told me: having a concrete project on your resume that touches NLP, even something small, makes the technical discussion round way easier. They want to see you've thought about language models in a real context, not just read about them.

If you're a new grad targeting Cohere in 2026, start prepping the OA early. That's the real bottleneck.

7 replies

newgrad_neil

this is incredibly helpful. the OA difficulty is what I was most unsure about. did both problems feel like standard LC or was there any Cohere-specific ML angle on the coding questions?

jp_newgrad

pure algorithms, no ML wrapper. the graph problem was about shortest paths (think BFS variation) and the DP one was a variation on sequence matching. nothing model-specific in the OA itself.

bootcamp_bri

I don't have a CS degree so the 'same bar as senior' comment stings a bit. do you know if they've hired anyone without a traditional background or is this strictly CS-heavy?

jp_newgrad

I can't say for certain. everyone I met in the process had CS degrees but that's a small sample. I'd say the OA is your honest filter since it's blind to your background. if you can get through it your resume comes after.

returner_ren

really appreciate you writing this out. the tip about having an NLP project on your resume is gold. even a sentiment classifier or something from a course project probably counts more than people think.

mobile_mara

75 minutes for two problems feels tight if the DP one was medium-hard. how was the pacing? like could you get partial credit or was it all-or-nothing test cases?

jp_newgrad

test-case based so partial passing matters. I passed maybe 8/10 on the DP problem and still moved forward. don't abandon a problem early if you have a working-ish solution.