Job Title: Dataset Management - Domain-Expert Product Engineer
Workplace: Remote / Full-time
Comepnsation: $220k - $250k
We are hiring on behalf of our client based in the United States for a Dataset Management - Domain-Expert Product Engineer.
Why this role exists
Our client's platform is where computer vision datasets live. Over a billion images, hundreds of thousands of projects, and the tools people use to label, curate, evaluate and improve them. The team's mission is simple to say and hard to do: help users get a CV model fast, then make it better.
The client has a strong engineering team. What they do not have is someone who knows dataset management itself better than any of them - who has spent years on data curation, embeddings, eval workflows and foundation models, and has opinions about where this category is going.
That is the hire. Not a pair of hands. A center of gravity.
What you'll work on
- What the team builds, for whom, and how they win. You will help shape DM product direction - bringing domain judgment the team doesn't have to decisions they're already making.
- Data curation at scale - sampling, deduplication, diversity, quality signals across a corpus measured in billions of images, where a naive query costs thousands of dollars.
- Evaluation - what "better" means for a dataset and for a model, and the tooling that proves it.
- Foundation models as a primitive - auto-labeling, embeddings, vector search, VLM-assisted workflows. The team ships against SAM 3, Cosmos, Gemini and whatever lands next month.
- Shipping. This is a full-stack engineering role on a small team. You will write production code the week you join.
What the client is looking for
- A top engineering bar, full-stack. You are comfortable from the database to the canvas.
- Real dataset-management depth. Embeddings and vector analysis, curation and data quality, eval workflows, working with foundation models. Not adjacent - the thing itself.
- Product judgment. You form opinions about competitors and positioning and can defend them.
- Self-direction. You are handed an area, not a ticket queue. You find the next step yourself and raise the ceiling of the people around you.
- Room to grow. The surface here is larger than any one person can hold. Responsibility is available to whoever earns it.
What the client is actually evaluating
Everyone on the team writes code with AI all day, so the interesting question was never whether you can type. Across the process, the client is looking for four things:
- Understanding - can you debug what you shipped, and explain why you chose it?
- Creativity - non-obvious approaches, not the first pattern that fits.
- Problem selection - of ten things you could do, which matters, and why.
- Analytical thinking - reasoning through ambiguity with incomplete data.
The team you'd join
Seven people on the client's Dataset Management team, remote, weighted toward Brazil. Small enough that your work is visible the day it ships. The team currently touches: the annotation and auto-labeling stack, the agent/MCP experience, active learning, video as a first-class primitive, workflow evaluation, and data partnerships with Meta and NVIDIA.