eBay's posting for an Applied Researcher on the NYC recommender systems team is the most revealing in the batch — it's a snapshot of how a legacy marketplace giant is fighting to stay relevant against Amazon's recommendation engine and Shopify's fragmented commerce model.

The stack tells a story of pragmatic scale:

• Spark/YARN — thousands of nodes, suggesting Hadoop-era infrastructure that's been kept alive rather than ripped out for Kubernetes/Ray. This is "we have petabytes and we're not migrating just for fashion."
• XGBoost alongside deep learning — eBay isn't drinking the all-neural-nets Kool-Aid. Gradient boosting still wins on tabular click data, and admitting that publicly is refreshing.
• Online learning — the interesting one. This signals real-time model updates, which matters when inventory is auction-driven and item lifetimes are measured in days, not the months Amazon enjoys with catalog SKUs.

What the posting reveals about stage and direction: The phrase "reinvent the recommender systems experience" combined with linking to a literal item page (ebay.com/itm/391756623227) is telling. They're not pitching greenfield work — they're pitching the chance to fix something visible and broken. That's a mature-company tell: the surface area is the product, and they know it's underperforming. The "true grit" and "most challenging codebases" line is HR-speak for "expect legacy Java and undocumented Scala jobs."

Skills/trends highlighted:

• Applied researchers who can also ship — "50% coding" is the giveaway. Pure-research PhDs need not apply; this is the ML engineer-researcher hybrid role that's eaten the industry since 2022.
• Willingness to take junior candidates for a recommender role is unusual and a green flag — suggests a senior-heavy team that needs hands, not a thin team needing a savior.
• NYC location for ML is deliberate: away from the SF talent war, closer to ad/finance ML talent pools.

Red flags: The "most challenging codebases and the most elegant systems alike" phrasing is a polite warning that you'll spend time in both. The terse posting and direct-email application ([email protected]) suggest a hiring manager bypassing the corporate funnel — which is great for response time but means onboarding may be lonely.

Green flags: VISA sponsorship, junior-friendly, real production scale, and an honest scope (improve this specific page) rather than vague "transform the business" language.