HubSpot · Primly Community

HubSpot machine learning engineer interview: what they actually tested, 2026

ml_mike · 5 replies

I interviewed for an MLE role at HubSpot earlier this year. Wanted to share what the process looked like because HubSpot ML isn't well-documented and I had basically zero info going in.

First thing to understand: HubSpot ML is mostly applied ML for their CRM/marketing products. Think email send-time optimization, lead scoring, deal probability, content recommendation. Not research. Not GenAI moonshots (well, there are some AI features now but the MLE roles I interviewed for were more classical). Keep that framing when you prep.

Recruiter screen. They asked about my ML background but spent time on product intuition too. "What ML models do you know that could help a sales rep prioritize their pipeline?" came up early. Have an answer.

Technical phone screen (90 min). Split: coding half + ML design half. Coding: Python, LeetCode-medium, nothing exotic. A sliding window problem. They cared about clean code and talking through tradeoffs. ML design: "Design a model to predict which leads are likely to convert within 30 days." They want you to walk through data, features, target definition, model choice, evaluation metrics, deployment/monitoring. I went with a gradient boosting approach (XGBoost), talked about class imbalance, precision vs recall tradeoff for the sales use case. They pushed on the metric choice specifically.

Onsite (5 rounds, virtual). 2x coding (same LeetCode-medium flavor, one had a data manipulation / pandas component) ML system design: I designed a content recommendation system for HubSpot's content hub product. Lots of discussion on cold-start, feedback loops, offline vs online evaluation. ML breadth: rapid-fire conceptual questions. Bias-variance, regularization, gradient boosting internals, recommendation system gotchas, a/b testing validity. Not gotchas, just depth checks. Behavioral: cross-functional work, influencing without authority, handling a model that underperformed in production.

Total: 4.5 weeks. Offer was around $195k base for senior level, Boston. RSU package was decent, not FAANG-tier but reasonable for the COL.

The ML bar felt solid but not punishing. Know your fundamentals and be able to reason about the product context.

5 replies

ds_dmitri

The metric choice deep-dive is real. I got the same kind of pushback in a DS interview there. They really want you to justify why precision vs recall for a specific business decision, not just say 'I'd use F1 score.' Prepare to defend.

staff_steph

The 'influencing without authority' behavioral question at HubSpot is almost universal across all the loops I've seen people describe. Something about their org structure makes cross-functional influence a real day-to-day thing. Prep a specific story with a clear outcome.

newgrad_neil

How much did they care about deep learning specifically? I have more experience with tree-based models than neural networks and I'm worried that's a gap.

ml_mike

Honestly, not much for this role. They're using tree-based and classical methods for most of their production models. Deep learning came up briefly in the ML breadth round but it wasn't a focus. Know the concepts, but you don't need to be a transformer researcher.

qa_quinn

Did they ask about model testing / quality in production at all? That part always gets skipped in these writeups but it's half the actual job.