Ciaran S.

Key Skills

Software

Data Annotation Tech

Other

Top Subject Matter

AI training model evaluation

prompt engineering

and quality assurance

Top Data Types

Image

Text

Video

Document

Top Task Types

Emotion Recognition

Evaluation/Rating

Prompt + Response Writing (SFT)

Text Summarization

Translation/Localization

Classification

RLHF

Fine-tuning

Red Teaming

Freelancer Overview

LLM evaluator and data annotator with experience contributing to AI training pipelines across preference labelling, output evaluation, red-teaming, and supervised fine-tuning (SFT) data generation. Background in legal analysis and capital markets provides specialist domain capability for high-stakes content evaluation.

Labeling Experience

Response Authoring and Gold-Standard Creation

TextFine-tuning

Editing model outputs to produce gold-standard responses meeting defined quality, accuracy, and style criteria; Authoring reference responses from scratch for supervised fine-tuning datasets; Applying domain expertise in legal analysis and capital markets to produce contextually accurate, appropriately hedged responses in specialist subject areas.

2025 - Present

Prompt Engineering and Training Data Generation

TextPrompt + Response Writing (SFT)

Designing complex, multi-turn prompts to probe model capabilities and surface edge case failures; Generating training prompts across capability dimensions including logical reasoning, factual recall, and calibrated uncertainty expression; Targeting specific failure modes in prompt design: instruction drift, sycophancy, context confusion, and refusal miscalibration.

2025 - Present

Model Output Evaluation and Quality Assurance

TextRLHF

Evaluating model responses against multi-dimensional rubrics covering factual accuracy, reasoning quality, instruction following, coherence, and alignment with intended behaviour; Assessing calibration and epistemic honesty: identifying overconfidence, hallucination, and failure to express appropriate uncertainty; Conducting comparative preference labelling: ranking competing model outputs against structured criteria to generate RLHF training signal.

2025 - Present

Red-Teaming and Safety Evaluation

Data Annotation TechTextRed Teaming

Conducting systematic adversarial testing to identify potential harms, biases, and safety policy violations in model outputs; Applying structured harm taxonomies to classify risk level and violation type across response categories; Evaluating model behaviour for compliance with content policies, ethical guidelines, and applicable legal standards; Identifying and flagging sycophancy, manipulative reasoning, and deceptive framing patterns.

2024 - Present

Education

Work History