Robson Caldeira

Key Skills

Software

Appen

Clickworker

CVAT

Labelbox

Lionbridge

OneForma

Remotasks

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Image

Text

Video

Top Task Types

Prompt + Response Writing (SFT)

Question Answering

Red Teaming

RLHF

Text Generation

Freelancer Overview

I am an AI Data Labeling and LLM Evaluation Specialist with over 1 year of experience in text, image, and video annotation. I have contributed to a variety of AI training projects, including LLM prompt evaluation, text generation, content moderation, and computer vision tasks. My background in data analytics and my fluency in both Portuguese (PT-BR) and English allow me to work on diverse datasets with high precision and cultural sensitivity. I am proficient in tools such as Labelbox, CVAT, and platforms like Outlier, Alignerr, Remotasks, and OneForma. My key strengths include prompt engineering, reinforcement learning with human feedback (RLHF), text summarization, and image/video labeling. I am passionate about delivering high-quality annotations to enhance AI model performance across industries.

Labeling Experience

LLM Evaluation and Prompt + Response Writing

LabelboxTextQuestion AnsweringText Generation

This project involves evaluating the performance of large language models by analyzing generated responses, assessing prompt completions, and writing new prompts and responses for supervised fine-tuning (SFT). Tasks include Red Teaming to identify vulnerabilities, Question Answering validation, and rating outputs according to detailed guidelines to enhance model quality. The work demands high attention to linguistic nuances, logical consistency, cultural context, and alignment with client-specific quality standards. All annotations undergo strict quality assurance processes to maintain project accuracy benchmarks.

This project involves evaluating the performance of large language models by analyzing generated responses, assessing prompt completions, and writing new prompts and responses for supervised fine-tuning (SFT). Tasks include Red Teaming to identify vulnerabilities, Question Answering validation, and rating outputs according to detailed guidelines to enhance model quality. The work demands high attention to linguistic nuances, logical consistency, cultural context, and alignment with client-specific quality standards. All annotations undergo strict quality assurance processes to maintain project accuracy benchmarks.

2024

Education

Work History