Video Data Reviewer - Egocentric - AI Trainer

Obsidian; Boston (MA); On-site; USD 25,000 - 33,000; Full time

Job summary

Obsidian is seeking 5 detail-oriented generalists to support a series of short-term projects involving egocentric (first-person) video data. The engagement runs for about 2 weeks across 3 projects, with each project lasting roughly 3 days to 1 week.

The first project starts Monday morning PST, August 17, to kick off this fast-paced data-review effort. You will review egocentric video data, applying project-specific guidelines with careful attention to detail.

Qualifications

• Excellent attention to detail and visual comprehension as core requirements.
• Ability to follow structured, project-specific guidelines precisely.
• Strong written English communication for notes and reports.

Responsibilities

• Review egocentric video data and annotate according to guidelines.
• Assess accuracy of action segment boundaries and adjust/flag as needed.
• Maintain consistency across many short segments and multiple videos.

Skills

Attention to detail

Visual comprehension

Follow guidelines

English communication

Tools

Video annotation tools

Job description

About the Role

We are looking for 5 detail-oriented generalists to support a series of short-term projects involving egocentric (first-person) video data. The engagement will run for approximately 2 weeks, covering 3 projects, with each individual project expected to last between 3 days and 1 week. The first project is expected to begin Monday morning 17th August (PST).

What You'll Do

You will review egocentric video data and complete structured quality-review tasks according to project-specific guidelines. The work requires strong attention to detail, consistency, and the ability to make precise judgments about actions occurring in first-person video.

Project 1: Action Segment Boundary Review

For the first project, you will review the boundaries of action segments within egocentric videos.

Dataset

• 50 egocentric videos

• Approximately 1-2 minutes per video

• Approximately 38 action segments per video on average

• Estimated review time: ~30 minutes per video

Responsibilities

• Review action segments identified within each video

• Determine whether segment start and end boundaries accurately correspond to the action being performed

• Adjust or flag incorrect boundaries according to provided guidelines

• Maintain consistent judgment across a high volume of short action segments

• Complete assigned videos within the project timeline

Additional egocentric-data projects will follow during the 2-week engagement. Detailed instructions will be provided at the start of each project.

Who We're Looking For

You may be a strong fit if you:

• Have excellent attention to detail and visual comprehension

• Can carefully distinguish between closely related actions and moments in video

• Are comfortable performing structured, repetitive review work while maintaining accuracy

• Can quickly learn and consistently apply detailed annotation guidelines

• Have strong written English communication skills

Prior experience with video annotation, data labeling, computer vision datasets, or egocentric video is helpful but not required.

Engagement Details

• Openings: 5

• Duration: Approximately 2 weeks

• Projects: 3 short-term projects

• Individual project duration: Approximately 3 days to 1 week

• First project start: Monday morning PST