Misael Corrales

Key Skills

Software

Labelbox

Remotasks

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Computer Code Programming

Image

Top Task Types

Audio Recording

Classification

Computer Programming Coding

Evaluation Rating

RLHF

Freelancer Overview

As a Freelance Data Annotator since September 2024, I’ve specialized in creating and refining AI training data for complex, real-world applications. I’ve developed 200+ high-quality prompts across video, voice, and scientific domains, enabling AI models to perform better in specialized and multimodal tasks. My work includes rigorous Likert-scale evaluations of AI responses—assessing critical dimensions like accuracy, safety, coherence, and helpfulness—and active participation in RLHF (Reinforcement Learning from Human Feedback) projects. In these projects, I provided detailed comparative analysis and qualitative feedback to enhance model outputs, with a sharp focus on factual integrity, logical consistency, and ethical alignment. What sets me apart is my blend of technical precision and cross-disciplinary insight. My MITx certification in Machine Learning with Python and IBM’s Prompt Engineering course arm me with structured methodologies for data curation and iterative feedback. Beyond this, I leverage my background in aerospace engineering and quality management to approach data labeling with exceptional rigor and foresight. Teaching university courses in AI/ML further deepens my understanding of how training data impacts real-world systems, allowing me to create datasets that prioritize not just performance but also reliability, safety, and domain-specif

Labeling Experience

Data Annotator

Scale AIAudioEvaluation Rating

As a reviewer in the Xylophone Panda Prompt Generation project, I contributed to training an AI speech model to generate natural, human-like conversational responses. My role involved creating and evaluating authentic audio prompts across diverse categories—such as high-quality dialogues, knowledge-based exchanges, and role-play scenarios—while ensuring each recording reflected real-world spontaneity (e.g., pauses, inflections, and background noise). I adhered to strict quality measures, including: Natural Delivery: Avoiding scripted tones and capturing genuine speech patterns (e.g., sighs, laughter, filler words like "um"). Contextual Alignment: Matching prompts to assigned conversation types and sub-categories (e.g., problem-solving or immersive role-play). Transcript Accuracy: Verifying transcripts mirrored audio exactly, including pauses marked with ellipses (…) and contractions (e.g., "I’m" vs. "I am").

As a reviewer in the Xylophone Panda Prompt Generation project, I contributed to training an AI speech model to generate natural, human-like conversational responses. My role involved creating and evaluating authentic audio prompts across diverse categories—such as high-quality dialogues, knowledge-based exchanges, and role-play scenarios—while ensuring each recording reflected real-world spontaneity (e.g., pauses, inflections, and background noise). I adhered to strict quality measures, including: Natural Delivery: Avoiding scripted tones and capturing genuine speech patterns (e.g., sighs, laughter, filler words like "um"). Contextual Alignment: Matching prompts to assigned conversation types and sub-categories (e.g., problem-solving or immersive role-play). Transcript Accuracy: Verifying transcripts mirrored audio exactly, including pauses marked with ellipses (…) and contractions (e.g., "I’m" vs. "I am").

2024

Data Annotator Generalist

Scale AIImageObject Detection

This project focused on evaluating and improving an AI Assistant’s ability to interpret and describe visual scenes accurately. The goal was to test the model’s comprehension of spatial relationships (e.g., front/back, left/right, top/bottom), object attributes (e.g., color, size, quantity), and contextual details within an image. The tasks were designed to mimic real-world user queries where visual understanding is critical—such as object counting, spatial reasoning, and detailed scene descriptions. Image Sourcing & Selection Prompt Design Evaluation & Feedback

This project focused on evaluating and improving an AI Assistant’s ability to interpret and describe visual scenes accurately. The goal was to test the model’s comprehension of spatial relationships (e.g., front/back, left/right, top/bottom), object attributes (e.g., color, size, quantity), and contextual details within an image. The tasks were designed to mimic real-world user queries where visual understanding is critical—such as object counting, spatial reasoning, and detailed scene descriptions. Image Sourcing & Selection Prompt Design Evaluation & Feedback

2024 - 2024

Math Expert anotator

Scale AIAudioEvaluation Rating

Evaluate and rewrite responses to provided math or reasoning prompts. The math level is expected to range from middle school to early college. The primary goal is to create responses that are more accurate and engaging than the existing model responses.

Evaluate and rewrite responses to provided math or reasoning prompts. The math level is expected to range from middle school to early college. The primary goal is to create responses that are more accurate and engaging than the existing model responses.

2024 - 2024

Education

Work History