Ariel S.

Key Skills

Software

Mindrift

OpenCV AI Kit (OAK)

Internal/Proprietary Tooling

Top Subject Matter

Industrial quality control—weld defect detection on car bodies

Medical image analysis—cataract and glaucoma diagnosis

Top Data Types

Image

Top Task Types

Classification

Object Detection

Freelancer Overview

As a Computer Science graduate with a background spanning machine learning research, full-stack development, and quality-critical industrial systems, I bring both technical depth and rigorous attention to detail to AI training data work. At Astra Daihatsu Motor, I developed an image recognition system for weld defect detection — a project that demanded precise visual judgment, consistent labeling standards, and an understanding of edge cases that challenge model generalization. This experience gives me a practitioner's perspective on what makes training data valuable. I am also a certified TensorFlow Developer (DeepLearning.AI) with additional credentials in machine learning and mathematics for ML, which means I understand not just how to label data, but why certain labels matter to model behavior. I work well in structured, detail-oriented environments and have demonstrated the ability to maintain quality standards under project deadlines — skills directly transferable to high-volume or specialized annotation tasks.

Labeling Experience

Rice Plant Disease Dataset

ImageClassification

As team lead for a machine learning product MVP, oversaw the data collection, cleaning, and labeling pipeline for our application's core model. Defined labeling guidelines for the team, reviewed annotations for consistency, and iterated on label taxonomy based on model performance feedback.

2023 - 2023

Medical Image Classification

ImageClassification

Worked with ophthalmological image datasets, applying Histogram of Oriented Gradients (HOG) feature extraction and preparing data for transfer learning with ResNet50V2.

2022 - 2022

Weld Defect Image Dataset

ImageClassification

As part of developing a CNN-based quality inspection system, I worked with industrial image data of car body welds, identifying and categorizing defect types (CO₂ weld defects) to build a labeled training dataset. This involved defining annotation criteria, handling ambiguous edge cases, and ensuring class consistency across thousands of images — directly informing the final ResNet50V2 model that replaced 5 human inspectors on the production line.

2022 - 2022

Education

Work History