Perplexity · Primly Community

Went through the full Perplexity ML eng loop last month. Here's the real breakdown.

mobile_mara · 4 replies

Applied cold via the website. Heard back in about 8 days, which felt fast. Here's the structure I went through:

Round 1: 30-min recruiter screen. Pretty standard, focused on motivation and timeline. She was direct and asked me to be direct too, which I appreciated.

Round 2: Technical with a senior ML eng. Not a leetcode grind. They gave me a real problem: design a ranking system for search results that handles freshness vs. relevance tradeoffs. Open-ended. They wanted me to reason through it, not solve it perfectly. I spent the first 15 min asking clarifying questions and they seemed to like that.

Round 3: Systems design + product intuition hybrid. This was the unusual one. Half was infra (how do you build low-latency retrieval at scale), half was basically "what would you change about how Perplexity answers this type of query." They want people who are users of the product, not just builders.

Round 4: Brief culture/values chat with someone on the leadership team. Light but not perfunctory. They probed for how I handle disagreement.

Total: 3.5 weeks from application to offer. Offer turnaround was 4 days after the last round. They do move fast.

One thing that surprised me: they explicitly said they were NOT looking for people who needed a lot of process overhead. If you need a lot of consensus before shipping, probably not the fit.

4 replies

infra_ines

the retrieval at scale question is interesting. did they go deep on vector DBs or more traditional inverted index territory? trying to figure out what to actually study

ml_mike

both, honestly. they didn't care about specific tools, they cared about tradeoffs. "when does dense retrieval beat sparse" type reasoning. i'd know why you'd pick HNSW over exact search rather than memorizing implementation details.

pm_priya

the "half systems half product intuition" hybrid round is so Perplexity. makes sense given how thin the line is between their eng and product work. useful datapoint.

corp_refugee

"not looking for people who need a lot of consensus before shipping" is every startup's way of saying the process is vibes-based and sometimes that's great and sometimes it's a nightmare. ask them for a specific example of a recent decision that was made fast and turned out wrong and how they handled it.