Ahimedsef F.

Key Skills

Software

CVAT

Label Studio

Labelbox

Scale AI

Other

Top Subject Matter

Technology – AI & Machine Learning

inguistics – NLP

Data Quality & Annotation

Top Data Types

Text

Image

Audio

Top Task Types

Bounding Box

RLHF

Transcription

Prompt + Response Writing (SFT)

Text Summarization

Data Collection

Evaluation/Rating

Question Answering

Freelancer Overview

I am a Native Amharic speaker (C2) with advanced English proficiency (C1) and a Google Data Analytics certification, specializing in RLHF, SFT, and AI safety evaluation for Amharic-language LLM training. I have a proven track record of 5,000+ annotated samples across text, image, audio, and video modalities with 97%+ accuracy on quality audits. I am an expert in pairwise comparison ranking, hallucination detection, harmfulness evaluation, and prompt-response generation for leading AI models. I have a deep understanding of Amharic linguistics, cultural nuance, and regional dialects, enabling precise evaluation of AI-generated content for fluency, factual accuracy, and cultural appropriateness. I am highly effective at following complex multi-page annotation guidelines, delivering consistent output under tight deadlines, and adapting rapidly to new task types. I am seeking to leverage my bilingual expertise and 1+ years of hands-on data annotation experience to improve AI model quality and safety.

Labeling Experience

Senior Data Annotator & RLHF Evaluator

TextRLHF

Completed 5,000+ annotation tasks across text, image, audio, and video modalities in Amharic and English, maintaining a 97%+ accuracy rate verified through multiple quality audit rounds. Performed RLHF-style pairwise comparisons and ranking evaluations on AI-generated Amharic responses, assessing helpfulness, harmlessness, factual accuracy, fluency, and cultural appropriateness. Conducted SFT data creation by writing high-quality Amharic prompt-response pairs and rewriting AI outputs to meet gold-standard reference quality. Executed hallucination detection and fact-checking on AI-generated Amharic content. Performed multimodal object tagging (bounding box, polygon, semantic segmentation). Managed 50-80 tasks per day with zero missed deadlines.

2024 - 2026

Education

Work History