Ibrahima Sidibe

Key Skills

Software

Appen

Labelbox

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Text

Video

Top Task Types

Action Recognition

Audio Recording

Evaluation Rating

Text Generation

Text Summarization

Freelancer Overview

I have word with LabelBox and Scale AI for 1,5 years on various projects in English and French regarding data labeling, filtering harmful content, rating response. On top of that, I was also assigned with generating prompt, analyzing voice recording and writing transcriptions from one language to another.

Labeling Experience

Voice evaluation

LabelboxAudioRLHFAudio Recording

On the software, we start with a voice recording according to the conditions set by the task. (quiet room, inside and certain tonality in the voice). It insist in a conversation with an AI for 5-10 min per task. At the end of the conversation, we rate the AI responses according to several criteria such as connection issues, instruction following, adaptation to the customs and also voice quality. Based on those information, we give a final rating to the AI

On the software, we start with a voice recording according to the conditions set by the task. (quiet room, inside and certain tonality in the voice). It insist in a conversation with an AI for 5-10 min per task. At the end of the conversation, we rate the AI responses according to several criteria such as connection issues, instruction following, adaptation to the customs and also voice quality. Based on those information, we give a final rating to the AI

2025 - 2025

LLM Evaluation

Scale AITextPrompt Response Writing SFT

The goal was to generate a prompt in a specific topic. From there, 2 repsonses are generated and rated based on several criterias ( verbosity, fluency, harmlessness but also truthfulness and instruction following). Based on that review, I give the prefered response. We had to do this for hundreds of subjects which each task specific to a certain subject.

The goal was to generate a prompt in a specific topic. From there, 2 repsonses are generated and rated based on several criterias ( verbosity, fluency, harmlessness but also truthfulness and instruction following). Based on that review, I give the prefered response. We had to do this for hundreds of subjects which each task specific to a certain subject.

2024 - 2025

Education

Work History