Amadeo Filberto Marli

Key Skills

Software

Labelbox

Mercor

OneForma

Remotasks

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Computer Code Programming

Image

Text

Top Task Types

Classification

Computer Programming Coding

Emotion Recognition

Evaluation Rating

Fine Tuning

Freelancer Overview

I have over a year of experience in AI training and evaluation. The most recent project I was involved in was Multilingual Static Comparison v3 at Scale AI (commissioned by Meta). As a top contributor, I evaluated high volumes of AI outputs in Indonesian, consistently ensuring accuracy, fluency, and contextual appropriateness. This work required not only bilingual proficiency, but also the ability to identify subtle linguistic differences and assess which outputs are best aligned with user expectations. Through this work, I contributed directly to improving RLHF pipelines and strengthening cross-lingual AI alignment. I bring proven expertise in data labeling, preference ranking, and linguistic quality control, with a strong record of reliability, precision, and scalability in AI evaluation workflows.

Labeling Experience

Contributor at Multilingual Static Comparison

Scale AITextRLHFFine Tuning

We aim to train LLMs that power an entertainment focused product where users interact informally, including but not limited to role-playing, flirting, trolling, humor, emotional support, and general assistance. The AI should embody a distinct voice, responding in ways that feel natural, dynamic, and fun, aligned with a character it represents. As trainers, we must correctly assess and give appropriate feedback when agents respond unaccordingly to their expected behaviors.

We aim to train LLMs that power an entertainment focused product where users interact informally, including but not limited to role-playing, flirting, trolling, humor, emotional support, and general assistance. The AI should embody a distinct voice, responding in ways that feel natural, dynamic, and fun, aligned with a character it represents. As trainers, we must correctly assess and give appropriate feedback when agents respond unaccordingly to their expected behaviors.

2024 - 2025

Education

Work History