Samuel Ngunjiri

Key Skills

Software

Appen

Clickworker

Data Annotation Tech

Mindrift

OneForma

Toloka

Internal/Proprietary Tooling

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Computer Code Programming

Text

Top Task Types

Classification

Computer Programming/Coding

Data Collection

Entity (NER) Classification

Freelancer Overview

I am a detail-oriented data labeling specialist with hands-on experience supporting AI model training through platforms such as Appen, OneForma, Mindrift, Clockworker, and Toloka. I have contributed to diverse projects, including LLM prompt evaluation, chatbot response rating, text classification, sentiment analysis, and image/text matching. With a professional background in financial data analysis and structured auditing, I bring a high level of precision, pattern recognition, and critical thinking to every task. My academic foundation in Mathematics and Computer Science, combined with years of experience working with structured data and implementing ERP systems, gives me a unique edge in handling text-heavy annotation tasks, structured data tagging, and business-domain AI applications. I excel in adhering to labeling guidelines, maintaining consistency across large datasets, and applying contextual judgment in nuanced evaluation scenarios.

Labeling Experience

Finance LLM Engineer & QA Specialist – Turing (Amazon Project)

Internal Proprietary ToolingTextClassificationText Generation

I worked on a large-scale AI training project with Turing for Amazon, where I specialized in developing and evaluating finance-related datasets. My role involved writing and refining complex prompts in areas such as financial statement analysis, valuation, tax compliance, and accounting standards. I also designed scoring rubrics and carried out detailed evaluations of AI-generated responses, ensuring that every answer met strict requirements for accuracy, completeness, and logical rigor. Beyond prompt creation, I contributed to error analysis and quality assurance, where I reviewed outputs, identified mistakes, and provided corrected solutions with clear step-by-step reasoning. I also prepared few-shot exemplars and tagged data with the right categories, which helped improve model performance and consistency. The work combined my background in mathematics, auditing, and finance with strong analytical writing skills, allowing me to create annotations that were both technically accurate

I worked on a large-scale AI training project with Turing for Amazon, where I specialized in developing and evaluating finance-related datasets. My role involved writing and refining complex prompts in areas such as financial statement analysis, valuation, tax compliance, and accounting standards. I also designed scoring rubrics and carried out detailed evaluations of AI-generated responses, ensuring that every answer met strict requirements for accuracy, completeness, and logical rigor. Beyond prompt creation, I contributed to error analysis and quality assurance, where I reviewed outputs, identified mistakes, and provided corrected solutions with clear step-by-step reasoning. I also prepared few-shot exemplars and tagged data with the right categories, which helped improve model performance and consistency. The work combined my background in mathematics, auditing, and finance with strong analytical writing skills, allowing me to create annotations that were both technically accurate

2025 - 2025

Geocoding and Address Validation for Mapping AI

TolokaTextGeocodingEvaluation Rating

Verified the accuracy of physical addresses by cross-checking information provided in prompts against linked web content or map locations. Tasks involved determining whether addresses matched real-world businesses or landmarks, classifying discrepancies (e.g., outdated, missing, incorrect), and flagging locations for further review. Maintained high precision while processing hundreds of listings under time-sensitive conditions.

Verified the accuracy of physical addresses by cross-checking information provided in prompts against linked web content or map locations. Tasks involved determining whether addresses matched real-world businesses or landmarks, classifying discrepancies (e.g., outdated, missing, incorrect), and flagging locations for further review. Maintained high precision while processing hundreds of listings under time-sensitive conditions.

2024 - 2024

LLM Prompt Evaluation and Response Rating

MindriftTextClassificationText Generation

Evaluated thousands of AI-generated text completions for appropriateness, coherence, factual accuracy, and tone alignment. Tasks included rating multiple responses to the same prompt, assessing alignment with user intent, and selecting best completions. Also contributed to supervised fine-tuning (SFT) by rewriting prompts/responses for model training. Worked under strict accuracy and consistency guidelines. Completed over 200+ rating sessions with >95% approval.

Evaluated thousands of AI-generated text completions for appropriateness, coherence, factual accuracy, and tone alignment. Tasks included rating multiple responses to the same prompt, assessing alignment with user intent, and selecting best completions. Also contributed to supervised fine-tuning (SFT) by rewriting prompts/responses for model training. Worked under strict accuracy and consistency guidelines. Completed over 200+ rating sessions with >95% approval.

2024 - 2024

Business and Finance Text Classification

AppenTextEntity Ner ClassificationClassification

Labeled short- and long-form business content including emails, reports, and chats for sentiment, topic classification, and NER tagging. Tasks involved identifying key entities (names, dates, currencies), flagging emotional tone, and verifying label consistency across batches. Maintained high QA scores while adhering to detailed taxonomy rules. Contributed to datasets used for customer service AI and financial LLMs.

Labeled short- and long-form business content including emails, reports, and chats for sentiment, topic classification, and NER tagging. Tasks involved identifying key entities (names, dates, currencies), flagging emotional tone, and verifying label consistency across batches. Maintained high QA scores while adhering to detailed taxonomy rules. Contributed to datasets used for customer service AI and financial LLMs.

2023 - 2023

Education

Work History