Staff Machine Learning Engineer

As a Staff Machine Learning Engineer at Babylist, you own personalization and decide where it goes. Millions of families depend on what we build.Agents write most of the code now. So the hard part is yours: what to model, how it should work and whether what shipped actually helped. You're still in the code for the genuinely hard problems — the embeddings, the ranking, the systems that don't exist yet. Agents handle the volume. You spend your time on the parts that need a person.

We’re building well beyond the registry now: the financial side of raising a kid, maternal health made simpler and more human, the education new parents are looking for and the community around them. Personalization runs across all of it — the homepage feed, what we recommend next, search — plus the platform underneath and the AI we already ship to families.

You’ll own a big piece of how personalization works, but you won’t be boxed into it. The roadmap is open. You have a say in which bets we make, and you pick what you take on next.

What You’ll Own

A Staff MLE here sets direction. You own personalization as a domain: the models behind the homepage feed, add-next recommendations and search personalization, and the foundational representations several teams build on. You set where it’s going over the next year or two, sequence the bets that get there and make the technical and product calls along the way. You’re still in the code. We don’t have architects who’ve stopped building.

If you stepped away, multiple teams would feel it. Your models show up on the surfaces millions of families use, and in the work other teams choose to build on top of them. You don’t need direct reports to have that reach. It comes from what you build.

In practice, you:

• Take a fuzzy business problem from the first sketch through to a production model, and stay on the hook for whether it actually helped customers.
• Build custom embeddings from raw data — domain‑specific representations that go well beyond off‑the‑shelf image and text models — and own them as several surfaces adopt them.
• Make the modeling and architecture calls that span teams and the ones that are expensive to reverse.
• Own the full lifecycle: orchestration, deployment, monitoring and the retraining loop that keeps a model honest in production.
• Set the standard for how personalization builds with AI. Decide what good looks like, build the evals that catch a model that’s confidently wrong before it ships.
• Partner with product, design and data as a peer, shaping what’s worth building from the start.
• Coach Senior engineers through the hard calls, the ambiguous ones as much as the technical ones.

A few problems people at this level are working on right now:

• Building the foundational embeddings that let every surface — feed, recommendations, search — personalize from one shared representation instead of each team rebuilding it.
• Resolving one customer across registry, shop and health, plus the friends and family buying for them, so recommendations work everywhere.
• Deciding what the registry recommends to each family, from the ranking to the model behind it, and proving in a live experiment that it actually helps.

Who You Are

You’ve shipped production ML for enough years to have earned strong opinions, and you hold them loosely. You can pick up an ambiguous problem and start moving before anyone hands you the full picture. You’ve already changed how a team builds with AI, and the new way stuck.

You’ve built recommender systems or personalization that reached real users at scale, and you can point to what moved because of it. You’re deep in the Python ML ecosystem (pandas, scikit‑learn, XGBoost, PyTorch) and fluent across the whole lifecycle, from orchestration to monitoring, not just training. The thing that sets you apart: you build custom representations from raw data instead of reaching for the off‑the‑shelf embedding.

A few things that tend to be true of people who thrive here:

• You measure yourself by impact: a customer outcome, or a model a dozen surfaces come to depend on.
• You go zero‑to‑one: define the problem space, architect from scratch and own it end to end.
• You’re curious: you spot problems before they’re filed and push your own ideas until they ship.

Compensation

We post real numbers. For a US‑based Staff Engineer, the starting base salary range is $233,500 to $290,700, plus a target annual bonus of 20 percent of base. That’s total target cash of roughly $280,200 to $348,840. On top of that you get meaningful equity and a 401(k) match. Where you start in that range depends on your experience, and your pay grows from there with performance and scope.

How We Build

AI is the default here. Engineers run agentic sessions for most of the work, and a lot of the interesting engineering now lives in the scaffolding that makes the agents good: the eval harnesses, the curated context, custom review skills and fast CI. Agents also triage incidents and handle a big share of support. A human always owns the outcome.

The architecture is intentionally simple: one Rails monolith, MySQL and few moving parts. That’s deliberate. Simple infrastructure lets us move fast and lets AI reason about the whole system, so the hardest problems are the ones in front of customers.

The Stack

• Core app: Rails, Packwerk, React, TypeScript, Sidekiq
• Mobile: iOS (Swift), Android (Kotlin)
• Machine learning: deep learning, matrix factorization, retrieval & ranking, AWS SageMaker, MLflow
• Key integrations: Shopify (payments), Iterable (CRM)

Why Babylist

An engineer, expecting her first baby, couldn’t find the registry she wanted. So she built it. That’s how Babylist started, and it’s still how we work: engineers solving problems for families.Becoming a parent is one of the biggest moments in a person’s life. Millions reach it for the first time every year, making thousands of decisions and figuring it out as they go. That’s who we build for, and we’re a long way from done.

Ten million people give gifts through Babylist every year. We did more than $750M in revenue in 2025, up 45 percent over the year before, and we’ve been profitable for eight years while staying independent. So you can take on a hard, multi‑year problem without watching over your shoulder for the next round or the next correction. And the team is small, around 65 engineers, so what you ship stays visible and your scope stays wide.

How We Work

Remote‑first across the US and Canada, and we have been for years. That’s not changing. We trust you to own your time and your outcomes, and we get everyone in a room together twice a year. Teams are small, pods of three to five engineers, so nothing you ship disappears into a committee.You’ll work shoulder to shoulder with product, design and data, and with the partners across the business who rely on what you ship. You’ll also stay close to customers yourself: sitting in on user interviews, watching session recordings, riding along with support. Here that’s part of the engineering job, on a regular basis.

How We Hire

Three rounds, usually two to three weeks start to finish.

• Recruiter conversation (30 minutes). Trade context: what you want next, what we’re building and straight answers on comp, team and remote