Key Skills

Software

Appen

Label Studio

Other

Internal/Proprietary Tooling

Data Annotation Tech

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Image

Text

Top Task Types

Action Recognition

Bounding Box

Classification

Data Collection

Emotion Recognition

Freelancer Overview

I am a LLM/NLP professional with 3 years as a data annotator, over 3 years in human translation and MTPE, and 1 year focused on (no‑code) prompt engineering and fine‑tuning. I design exhaustive critrions prompt/ rubrics, run multilingual audio/text annotation and LQA, and execute few‑shot/one‑shot, role‑play and comparative model evaluations to produce production‑ready datasets and improve model reliability; thats whats set me apart from others. I work with no‑code fine‑tuning tools and common annotation platforms (Label Studio and any other intern platform) and deliver measurable improvements in annotation quality and models performance.

Labeling Experience

Prompt Engineer - Rubrics Design

Data Annotation TechTextText GenerationFine Tuning

As a basis, a user prompt is given, along with three generic responses to that prompt. After generating the generic responses, I select the best one and, if editing is necessary, I edit it as a golden response to the user prompt. I then classify the three responses as poor, good, or excellent. After classification, the process of creating atomic and exhaustive criteria for a perfect rubric begins so that the model understands what the user expects as a response to the prompt. Test several times if the criterion affects the response for better performance and if it is a self-contained and atomic criterion for each part of the prompt, if it follows the classification, for example: role-play, summary, paraphrase instructions to follow, list, etc. And if the objective/subjective, implicit/explicit criteria are followed, as well as classifying and detecting what type of intent the prompt vs. response has. After carefully reviewing and confirming that all rubric criteria are complete without

As a basis, a user prompt is given, along with three generic responses to that prompt. After generating the generic responses, I select the best one and, if editing is necessary, I edit it as a golden response to the user prompt. I then classify the three responses as poor, good, or excellent. After classification, the process of creating atomic and exhaustive criteria for a perfect rubric begins so that the model understands what the user expects as a response to the prompt. Test several times if the criterion affects the response for better performance and if it is a self-contained and atomic criterion for each part of the prompt, if it follows the classification, for example: role-play, summary, paraphrase instructions to follow, list, etc. And if the objective/subjective, implicit/explicit criteria are followed, as well as classifying and detecting what type of intent the prompt vs. response has. After carefully reviewing and confirming that all rubric criteria are complete without

2025 - 2025

TrustScale

Internal Proprietary ToolingTextSegmentationClassification

Identifing in the AI model's response according to the user's query whether it is true and accurate or whether it is a hallucination. Note core claims in the responses and conduct in-depth research to find reliable sources that indicate whether it is true or not.

Identifing in the AI model's response according to the user's query whether it is true and accurate or whether it is a hallucination. Note core claims in the responses and conduct in-depth research to find reliable sources that indicate whether it is true or not.

2025 - 2025

Appen - Jigglypuff Project

AppenAudioBounding BoxSegmentation

Identification of speakers (speaker1/speaker2) and identification of accents according to the source language (Portuguese in this project). After identifying the speakers and accents, I transcribed the entire audio recording.

Identification of speakers (speaker1/speaker2) and identification of accents according to the source language (Portuguese in this project). After identifying the speakers and accents, I transcribed the entire audio recording.

2025 - 2025

Appen - Nidoking Project

AppenAudioBounding BoxSegmentation

The project involves first identifying the regions in which each speaker speaks and separating them into parts (no more than 25 seconds each), identifying background noise and sounds, labeling paused speech and filled pauses, and after all labeling, classification, and correction, transcribing into the corresponding language according to the guidelines.

The project involves first identifying the regions in which each speaker speaks and separating them into parts (no more than 25 seconds each), identifying background noise and sounds, labeling paused speech and filled pauses, and after all labeling, classification, and correction, transcribing into the corresponding language according to the guidelines.

2025 - 2025

Education

Work History