Ewalefoh Prosper

Key Skills

Software

Appen

CVAT

Labelbox

OneForma

Telus

Top Subject Matter

No subject matter listed

Top Data Types

Image

Text

Video

Top Task Types

Bounding Box

Classification

Evaluation/Rating

Object Detection

Text Summarization

Freelancer Overview

An experienced data labeling and AI training specialist with over 3 years of experience, I have contributed to a wide range of machine learning projects across industries, including autonomous vehicles, natural language processing (NLP), and medical imaging. What sets me apart is my commitment to data quality and precision. I bring a strong background in linguistics and an in-depth understanding of annotation guidelines, enabling me to adapt to diverse project requirements and languages. My work on NLP projects includes summarization and entity recognition for large language models (LLMs), where I specialize in multilingual datasets. Combined with my ability to lead quality assurance efforts, train teams, and streamline workflows, I deliver data that empowers AI models to perform at their best.

Labeling Experience

VCG - Image Evaluation

OneformaImageEvaluation Rating

VCG (Visual Content Generation) is an artificial intelligence (AI) technology to create images, videos graphics, and other visual media. The images generated based on the prompt are called output data. There can be one or more images resulting from the prompt, and they come in various styles depending on the project. 1. Animation 2. Illustration 3. Sketch 4. Emoji Evaluation is the critical part of each task and its main purpose. The output images are being evaluated on two levels. First, each output image needs to be evaluated independently. Then, each image must be compared with the other one and ranked by preference. Hence, the main types of evaluation used across the VCG projects are: 1. Rating Scale 2. Ranking (side-by-side comparison) Using both rating scales and preference ranking together provides a comprehensive evaluation framework.

VCG (Visual Content Generation) is an artificial intelligence (AI) technology to create images, videos graphics, and other visual media. The images generated based on the prompt are called output data. There can be one or more images resulting from the prompt, and they come in various styles depending on the project. 1. Animation 2. Illustration 3. Sketch 4. Emoji Evaluation is the critical part of each task and its main purpose. The output images are being evaluated on two levels. First, each output image needs to be evaluated independently. Then, each image must be compared with the other one and ranked by preference. Hence, the main types of evaluation used across the VCG projects are: 1. Rating Scale 2. Ranking (side-by-side comparison) Using both rating scales and preference ranking together provides a comprehensive evaluation framework.

2024 - 2024

Text Summarization

OneformaTextText Summarization

Task Overview The project involves evaluating the ‘summarization' responses generated by a virtual assistant (assistant). The evaluation process involves reviewing an instruction, original input text (referred to as the "original", or "original text") and the assistant’s subsequent response(s). The instruction contains requirements for the summary, and the input text is the content to be summarized. This original text may take various forms, such as a single email, a series of emails in a thread, one-on-one or group chat messages, notifications, or other types of general text. Workflow 1. Instruction and Original Text Examination Thoroughly examine the Instruction and Original Text. 2. Original Text Assessment Assess and flag if Original Text contains any of the specified harmful/safety categories. 3. Response Evaluation Assess and evaluate all available responses based on the various evaluation criteria 4. Response Ranking 5. Provide Comments

Task Overview The project involves evaluating the ‘summarization' responses generated by a virtual assistant (assistant). The evaluation process involves reviewing an instruction, original input text (referred to as the "original", or "original text") and the assistant’s subsequent response(s). The instruction contains requirements for the summary, and the input text is the content to be summarized. This original text may take various forms, such as a single email, a series of emails in a thread, one-on-one or group chat messages, notifications, or other types of general text. Workflow 1. Instruction and Original Text Examination Thoroughly examine the Instruction and Original Text. 2. Original Text Assessment Assess and flag if Original Text contains any of the specified harmful/safety categories. 3. Response Evaluation Assess and evaluate all available responses based on the various evaluation criteria 4. Response Ranking 5. Provide Comments

2024 - 2024

Annotation

OneformaImageEvaluation Rating

The goal of the projects is to evaluate the quality of machine-generated products in relevance to an input request or prompt. The products can include texts, sketches, emojis, and more. Evaluation Guidelines: Flags The following output image flags was applied (if applicable). A description of each flag and when to apply it can be found on the Image Evaluation Guidelines. - Inappropriate - Sensitive - Stereotype Grading Dimensions Evaluate each output image on the following dimensions. Guidelines. - Structural Integrity - Input/Output Alignment • Text to Image Alignment • Image to Image Alignment Ranking Scale Compare and rank each Output Image based on the grades i assigned for the following dimensions found on the Image Evaluation Guidelines. - Structural Integrity - Input/Output Alignment • Text to Image Alignment • Image to Image Alignment Leave a Comment Provided a concise, specific, well-structured comment that explains your decision-making process to the engineers

The goal of the projects is to evaluate the quality of machine-generated products in relevance to an input request or prompt. The products can include texts, sketches, emojis, and more. Evaluation Guidelines: Flags The following output image flags was applied (if applicable). A description of each flag and when to apply it can be found on the Image Evaluation Guidelines. - Inappropriate - Sensitive - Stereotype Grading Dimensions Evaluate each output image on the following dimensions. Guidelines. - Structural Integrity - Input/Output Alignment • Text to Image Alignment • Image to Image Alignment Ranking Scale Compare and rank each Output Image based on the grades i assigned for the following dimensions found on the Image Evaluation Guidelines. - Structural Integrity - Input/Output Alignment • Text to Image Alignment • Image to Image Alignment Leave a Comment Provided a concise, specific, well-structured comment that explains your decision-making process to the engineers

2024 - 2024

Education

Work History