Mohammed Alsalman

Key Skills

Software

CrowdSource

Data Annotation Tech

Google Cloud Vertex AI

Labelbox

Surge AI

Other

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Image

Text

Top Task Types

Audio Recording

Classification

Evaluation Rating

RLHF

Text Generation

Freelancer Overview

Over the past three years I’ve specialized in creating, labeling, and validating high-impact datasets for large language models in Swedish, Arabic, and English. As an AI Trainer I have built corpus-level supervision signals for general-purpose Swedish LLMs and for a niche mathematical-reasoning model, rating outputs for fluency, factuality, harmfulness, and cultural localization. I also served as a Swedish map evaluator, applying geospatial QA guidelines to improve search and routing outputs. Beyond text, I have led multimodal labeling sprints—classifying imagery, video scenes, and environmental audio, as well as coordinating native-speaker audio collection—to ensure balanced, production-ready assets. What sets me apart is the blend of linguistic range and analytical rigor I bring to data quality. A background in economics trains my eye for statistical consistency, while professional translation work sharpens my sense of nuance across cultures. Coupled with hands-on experience running end-to-end annotation pipelines and delivering client-specific datasets, I consistently raise model accuracy and reduce error rates for downstream teams.

Labeling Experience

Outlier

Scale AITextRLHFEvaluation Rating

I was part of a Reinforcement-Learning-from-Human-Feedback (RLHF) effort to teach a general-purpose LLM to solve mathematics problems posed in Swedish. My work covered the full RLHF loop: (1) authoring and curating ~10 k prompt/solution pairs spanning basic arithmetic through university-level calculus and linear algebra; (2) rating model answers for logical correctness, step-by-step reasoning, clarity, and linguistic fluency; and (3) writing high-quality reference solutions that were used to train the reward model and perform SFT (supervised fine-tuning).

I was part of a Reinforcement-Learning-from-Human-Feedback (RLHF) effort to teach a general-purpose LLM to solve mathematics problems posed in Swedish. My work covered the full RLHF loop: (1) authoring and curating ~10 k prompt/solution pairs spanning basic arithmetic through university-level calculus and linear algebra; (2) rating model answers for logical correctness, step-by-step reasoning, clarity, and linguistic fluency; and (3) writing high-quality reference solutions that were used to train the reward model and perform SFT (supervised fine-tuning).

2024 - 2024

Peroptyx

OtherGeospatial Tiled ImageryClassificationMapping

Can't disclose

Can't disclose

2022 - 2023

Education

Work History