HumanSignal Labs projects span the full range of work needed to train and evaluate modern AI models. The projects available to you at any given time depend on client demand, your location, your languages, and your skills and expertise — so the mix changes over time. Here are the main types of work you may see as a contributor.
Data annotation and labeling
The most common category. You review data and add structured labels or judgments, working in Label Studio. Examples include:
- Image and video annotation (drawing bounding boxes, classifying content, rating quality)
- Audio annotation and transcription
- Text classification, tagging, and quality review
Model evaluation and AI training
Projects where you help improve AI models directly by judging their outputs:
- Rating and ranking AI-generated responses
- Writing or refining prompts
- Comparing model outputs against guidelines (evaluations, RLHF-style feedback)
- Red teaming — probing models to find weaknesses or unsafe behavior
Data collection and creation
Some projects involve creating brand-new data rather than labeling existing data — for example, recording audio, capturing photos or video, or participating in structured data-collection studies. Most of this work is remote, but some projects run at in-person collection sites and are advertised separately with their own eligibility requirements.
Expert and specialized projects
Projects that need domain expertise — such as medicine, law, finance, STEM fields, or software development — or specific language skills. These typically pay higher rates and may require verification of your credentials or a more involved qualification process. Keeping the skills, languages, and expertise on your profile up to date is the best way to be matched to them.
A note on availability
Work availability is not guaranteed and varies from week to week. When you're accepted to a project, you'll always see the pay rate and requirements before you begin. If your dashboard is empty right now, see "I don't see any projects or tasks yet" for what to check.