Lj Abergas

Key Skills

Software

Roboflow

Top Subject Matter

No subject matter listed

Top Data Types

Video

Top Label Types

Segmentation

Tracking

Freelancer Overview

As a detail-oriented Quality Assurance professional, I specialize in high-precision video annotation and motion tracking for AI training datasets. My experience centers on executing complex data labeling tasks where accuracy is paramount, specifically identifying tracker misalignments and ensuring precise action descriptions. I have a proven track record of maintaining high data integrity by documenting technical discrepancies and refining workflows to minimize error rates in fast-paced production environments. In my recent role at Atlas Capture, I served as the final quality checkpoint, ensuring that all data met rigorous internal quality standards and project-specific requirements before final acceptance. My background as a Subject Matter Expert has equipped me with a "quality-first" perspective, allowing me to bridge the gap between technical annotation and high-level project goals. I am adept at utilizing QA audits to identify subtle inconsistencies, making me a reliable asset for projects requiring flawless data validation.

Labeling Experience

Atlas Capture

RoboflowVideoSegmentationTracking

It's a cutting-edge initiative focused on building the foundational data layer for Physical AI. The goal is to teach AI models how to perceive and interact with the physical world by analyzing human movement. Atomic Action Labeling: Breaking down complex human tasks (like cooking or cleaning) into "atomic" or tiny, distinct actions. Dense Labeling: Unlike standard tagging, this requires a continuous, high-precision stream of labels where every hand-object interaction is captured in the correct order with zero gaps. Egocentric Perspective: Much of the data involves first-person (POV) video, which is critical for training robots or wearable AI to understand tasks from a human's point of view.

It's a cutting-edge initiative focused on building the foundational data layer for Physical AI. The goal is to teach AI models how to perceive and interact with the physical world by analyzing human movement. Atomic Action Labeling: Breaking down complex human tasks (like cooking or cleaning) into "atomic" or tiny, distinct actions. Dense Labeling: Unlike standard tagging, this requires a continuous, high-precision stream of labels where every hand-object interaction is captured in the correct order with zero gaps. Egocentric Perspective: Much of the data involves first-person (POV) video, which is critical for training robots or wearable AI to understand tasks from a human's point of view.

2025

Education

Work History