Roland Larsson Forsberg

Key Skills

Software

OneForma

Other

Internal/Proprietary Tooling

Top Subject Matter

No subject matter listed

Top Data Types

Document

Image

Text

Top Task Types

Prompt + Response Writing (SFT)

RLHF

Translation/Localization

Freelancer Overview

I have extensive hands-on experience in AI training and data labeling through projects with platforms such as Outlier AI and OneForma. My work includes high-precision evaluation of both language and vision models, detailed rubric-based scoring, multimodal content assessment, and strict guideline application across complex tasks. I specialize in analyzing model behavior, detecting quality issues, and ensuring that outputs meet standards for accuracy, safety, localization, and user intent alignment. I have worked on tasks involving text classification, content safety review, translation and localization QA (sv_SE ↔ EN), prompt evaluation, image–text consistency checks, and long-context reasoning analysis. My background gives me a strong ability to follow instructions, identify subtle linguistic or logical errors, and provide consistent, high-quality annotations. I am reliable, detail-oriented, and able to work efficiently in environments with strict deadlines and evolving project guidelines.

Labeling Experience

Prompt Evaluation, Safety Review & Image–Text Consistency

OneformaImageQuestion AnsweringObject Detection

Evaluated prompts and responses for multimodal tasks. Labeled safety risks, checked image–text consistency, corrected model hallucinations, and reviewed user intent. Performed SFT-style improvements of prompt clarity and response quality.

Evaluated prompts and responses for multimodal tasks. Labeled safety risks, checked image–text consistency, corrected model hallucinations, and reviewed user intent. Performed SFT-style improvements of prompt clarity and response quality.

2025 - 2025

Swedish Localization, Translation QA & RLHF Linguistic Rating

OneformaTextClassificationText Generation

Reviewed and rated Swedish–English translations with RLHF-style linguistic scoring. Assessed tone, correctness, units, idioms, formatting, and cultural relevance. Corrected misuse, improved clarity, and validated guideline compliance.

Reviewed and rated Swedish–English translations with RLHF-style linguistic scoring. Assessed tone, correctness, units, idioms, formatting, and cultural relevance. Corrected misuse, improved clarity, and validated guideline compliance.

2025 - 2025

Multimodal RLHF: Image–Text Alignment & Safety

Internal Proprietary ToolingImageClassificationQuestion Answering

Evaluated image–text consistency in multimodal RLHF tasks. Labeled hallucinations, correctness, object presence, OCR errors, and unsafe content. Ranked and rated visual QA responses using Outlier’s rubric system for multimodal RLHF alignment.

Evaluated image–text consistency in multimodal RLHF tasks. Labeled hallucinations, correctness, object presence, OCR errors, and unsafe content. Ranked and rated visual QA responses using Outlier’s rubric system for multimodal RLHF alignment.

2025 - 2025

RLHF Evaluation – LLM Response Rating

Internal Proprietary ToolingTextText GenerationText Summarization

Performed RLHF-style evaluation of AI model responses, including instruction-following checks, truthfulness analysis, safety classification, tone assessment, and detailed rubric scoring. Ranked responses, corrected outputs, analyzed prompts, and contributed to SFT-style improvements.

Performed RLHF-style evaluation of AI model responses, including instruction-following checks, truthfulness analysis, safety classification, tone assessment, and detailed rubric scoring. Ranked responses, corrected outputs, analyzed prompts, and contributed to SFT-style improvements.

2025 - 2025

Education

Work History