Mark Daley

Key Skills

Software

AWS SageMaker

Argilla

CVAT

Google Cloud Vertex AI

Kili Technology

Labelbox

LabelImg

Label Studio

OpenCV AI Kit (OAK)

Redbrick AI

Roboflow

V7 Labs

Other

Internal/Proprietary Tooling

SuperAnnotate

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Text

Video

Top Task Types

Classification

Entity Ner Classification

Fine Tuning

Prompt Response Writing SFT

RLHF

Freelancer Overview

I am an experienced AI data labeler and LLM training specialist with a strong software engineering background. Over the past several years I’ve worked on projects involving text, documents, and source code, creating high-quality training data for large language models and evaluation pipelines. My work ranges from fine-grained classification and entity labeling to complex prompt–response writing, red-teaming, and human evaluation (ELO/rating) for both general-purpose and domain-specific models. I’m comfortable following detailed guidelines, thinking critically about edge cases, and providing clear written rationale for my decisions. Because I also write and review production code, I’m especially strong on code understanding, debugging and explaining code in natural language, and designing prompts that test reasoning and robustness. I’m reliable, fast, and very quality-driven, and I take pride in producing consistent annotations that model developers can trust.

Labeling Experience

Video Object Detection and Tracking with SuperAnnotate

SuperannotateVideoBounding BoxPolygon

I worked on a video annotation project using SuperAnnotate to create high-quality training data for computer vision and multimodal AI models. The work included frame-by-frame bounding box and polygon annotations for people and objects, assigning and maintaining tracking IDs across frames, and labeling actions or scene-level classes where required. I annotated and reviewed thousands of frames and short clips, followed strict project guidelines, and used built-in QA workflows in SuperAnnotate to ensure consistency and accuracy. The resulting dataset was used to improve object detection, tracking robustness, and downstream video understanding tasks.

I worked on a video annotation project using SuperAnnotate to create high-quality training data for computer vision and multimodal AI models. The work included frame-by-frame bounding box and polygon annotations for people and objects, assigning and maintaining tracking IDs across frames, and labeling actions or scene-level classes where required. I annotated and reviewed thousands of frames and short clips, followed strict project guidelines, and used built-in QA workflows in SuperAnnotate to ensure consistency and accuracy. The resulting dataset was used to improve object detection, tracking robustness, and downstream video understanding tasks.

2025 - 2025

LLM Evaluation and Prompt–Response Data Labeling (Text & Code)

Aws SagemakerComputer Code ProgrammingQuestion AnsweringText Generation

I worked on a large-scale LLM training and evaluation project focused on English text and source code. My tasks included writing and reviewing prompts, creating high-quality reference answers, and rating model responses on correctness, reasoning depth, style, and safety. For code-related items, I evaluated solutions, explained bugs in natural language, and wrote improved versions of code and explanations. The project covered several tens of thousands of examples and required strict adherence to detailed guidelines, careful handling of edge cases, and consistent rationales so that the data could be used directly for supervised fine-tuning and RLHF.

I worked on a large-scale LLM training and evaluation project focused on English text and source code. My tasks included writing and reviewing prompts, creating high-quality reference answers, and rating model responses on correctness, reasoning depth, style, and safety. For code-related items, I evaluated solutions, explained bugs in natural language, and wrote improved versions of code and explanations. The project covered several tens of thousands of examples and required strict adherence to detailed guidelines, careful handling of edge cases, and consistent rationales so that the data could be used directly for supervised fine-tuning and RLHF.

2024 - 2024

Education

Work History