Data Labeling Operations Manager

Bobyard•San Francisco, CA

1d•$90,000 - $125,000•Onsite

About The Position

Bobyard is building the AI that brings visual intelligence to construction. We're a Series A startup backed by 8VC, Primary, and Pear, and our models are trained on millions of construction drawings to help contractors estimate and bid faster. We're small, moving fast, and the work we ship directly changes whether a contractor wins or loses a bid. Our models are only as good as the data behind them. You'll own the labeling operation end to end — the annotator team, the quality bar, the datasets themselves. This is a first-in-function hire: there's no playbook waiting for you, you'll build it. Success looks like a labeling operation that's fast, accurate, and trusted enough that ML engineers stop double-checking your team's work.

Requirements

Direct experience managing a labeling, annotation, or data-quality team
Extremely detail-oriented — you notice when data is wrong, inconsistent, or incomplete before anyone points it out
Strong operational instincts — you can run many datasets, annotators, and priorities at once without dropping the details
Technical enough to work with ML engineers — you understand false positives, false negatives, class imbalance, and train/test splits, and you can set up your own tools to speed up labeling
Resourceful — when we need a new kind of data, you figure out how to find it
High ownership — you don't just coordinate the work, you make sure the dataset is actually good

Nice To Haves

Familiarity with labeling platforms like Labelbox, CVAT, or Supervisely
Basic SQL or Python for querying and cleaning data
Background in construction, CAD, or other visually complex technical domains

Responsibilities

Build and run the annotator team — recruit, onboard, train, and hold the bar on quality and throughput
Own labeling quality — review annotations, catch systematic errors before they hit a model, and turn what you find into sharper guidelines
Clean up the datasets we already have — fix inconsistent labels, missing metadata, duplicates, and other issues quietly hurting model performance
Source new data — find and organize construction drawings that expand our coverage of formats, classes, and edge cases we're currently missing
Turn ML requests into shipped datasets — scope the ask, run the project, deliver clean data on time
Work directly with ML engineers to understand where models are failing and build the data that fixes it
Build the tooling and workflows that make labeling faster and more reliable — this isn't just people management, it's systems work