Hosea Stephen

Key Skills

Software

Appen

Labelbox

Mindrift

OneForma

Internal/Proprietary Tooling

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Geospatial Tiled Imagery

Image

Text

Video

Top Task Types

Action Recognition

Bounding Box

Classification

Entity Ner Classification

Evaluation Rating

Geocoding

Mapping

Prompt Response Writing SFT

Question Answering

Routing

Text Generation

Text Summarization

Translation Localization

Freelancer Overview

I have a strong foundation in engineering, software development, and product delivery, with hands-on experience managing projects that require meticulous attention to data quality and process compliance. My background includes leading cross-functional teams to deliver digital platforms, where I coordinated backend development in Python and TypeScript, conducted user acceptance testing, and ensured robust documentation and data-driven decision-making throughout the product lifecycle. I am skilled in Python, SQL, cloud platforms (AWS, Azure), and have practical experience with data processing and automation using tools like OpenCV and Tesseract, which are directly applicable to data labeling and AI training data workflows. My experience working in regulated environments and performing detailed case analysis has strengthened my ability to ensure data accuracy, consistency, and compliance—key aspects for high-quality AI model training.

Labeling Experience

Project Coffee - Generative AI Text Evaluation

AppenTextClassificationEvaluation Rating

Evaluated the quality, safety, and coherence of text generated by Large Language Models (LLMs) to improve conversational AI capabilities. Conducted side-by-side (SXS) comparisons of AI responses, rating them based on helpfulness, honesty, and harmlessness (HHH framework). Identified and annotated hallucinations, factual errors, and logical inconsistencies in complex query responses. Authored high-quality "Golden Set" rewrites to train models on preferred prose styles and reasoning patterns.

Evaluated the quality, safety, and coherence of text generated by Large Language Models (LLMs) to improve conversational AI capabilities. Conducted side-by-side (SXS) comparisons of AI responses, rating them based on helpfulness, honesty, and harmlessness (HHH framework). Identified and annotated hallucinations, factual errors, and logical inconsistencies in complex query responses. Authored high-quality "Golden Set" rewrites to train models on preferred prose styles and reasoning patterns.

2025 - 2025

Project Latte - Multimodal Image Dialogue & Localizatio

Internal Proprietary ToolingImageBounding BoxClassification

Generated natural language descriptions and dialogue pairs based on visual inputs to train Multimodal Large Language Models (MLLMs). Assessed image captions for cultural relevance and localized nuance, ensuring descriptions were appropriate for specific target regions. Performed rigorous Quality Assurance (QA) on machine-translated content, correcting grammatical errors and ensuring "native-level" fluency. Verified visual grounding by checking if AI-generated text accurately reflected objects and spatial relationships present in the images.

Generated natural language descriptions and dialogue pairs based on visual inputs to train Multimodal Large Language Models (MLLMs). Assessed image captions for cultural relevance and localized nuance, ensuring descriptions were appropriate for specific target regions. Performed rigorous Quality Assurance (QA) on machine-translated content, correcting grammatical errors and ensuring "native-level" fluency. Verified visual grounding by checking if AI-generated text accurately reflected objects and spatial relationships present in the images.

2023 - 2024

Search Relevance & Maps

Internal Proprietary ToolingGeospatial Tiled ImageryEntity Ner ClassificationGeocoding

Analysed and verified map data accuracy, including business names, addresses, and hours of operation, to improve digital map services. Evaluated search query results for relevance and user intent, distinguishing between navigational, informational, and transactional queries. Investigated and corrected routing logic issues, ensuring turn-by-turn navigation data was safe and legally compliant. Applied local cultural knowledge to assess the relevance of "near me" search results, significantly improving the user experience for local queries

Analysed and verified map data accuracy, including business names, addresses, and hours of operation, to improve digital map services. Evaluated search query results for relevance and user intent, distinguishing between navigational, informational, and transactional queries. Investigated and corrected routing logic issues, ensuring turn-by-turn navigation data was safe and legally compliant. Applied local cultural knowledge to assess the relevance of "near me" search results, significantly improving the user experience for local queries

2022 - 2023

Education

Work History