Morris Murrah

Key Skills

Software

CrowdSource

CVAT

Data Annotation Tech

Deep Systems

Labelbox

Remotasks

Scale AI

Toloka

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Computer Code Programming

Document

Geospatial Tiled Imagery

Image

Text

Video

Top Label Types

Audio Recording

Evaluation Rating

Fine Tuning

Geocoding

Mapping

Polygon

Prompt Response Writing SFT

Question Answering

Red Teaming

RLHF

Freelancer Overview

I am an experienced AI and data annotation specialist with a strong background in training, evaluating, and optimizing large language and multimodal models. My work spans diverse domains, including legal reasoning, natural language processing, and computer vision, where I have designed and executed high-precision data labeling, prompt engineering, and rubric-based evaluation for enterprise AI systems. I excel at identifying edge cases, ensuring data quality, and aligning AI outputs with safety, accuracy, and ethical standards. Through hands-on roles in projects like legal AI model training and visual asset annotation, I have developed expertise in detailed guideline development, error analysis, and quality assurance at scale. My approach emphasizes cross-functional collaboration and continuous improvement to deliver reliable, real-world AI solutions.

Labeling Experience

Image labelling

CVATImagePrompt Response Writing SFT

I managed the end-to-end data annotation process for a large-scale computer vision project aimed at improving object detection models. My primary responsibility involved manually labeling over 5,000 high-resolution images, specifically identifying and categorizing diverse objects using bounding boxes and polygons. To ensure high data integrity, I utilized professional labeling software to maintain a consistent output. I strictly adhered to a 98% accuracy threshold, performing regular self-audits and cross-referencing against project-specific guidelines to eliminate labeling bias and errors. Throughout the project, I collaborated with the data science team to refine the labeling taxonomy, ensuring the final dataset was optimized for model training and deployment.

I managed the end-to-end data annotation process for a large-scale computer vision project aimed at improving object detection models. My primary responsibility involved manually labeling over 5,000 high-resolution images, specifically identifying and categorizing diverse objects using bounding boxes and polygons. To ensure high data integrity, I utilized professional labeling software to maintain a consistent output. I strictly adhered to a 98% accuracy threshold, performing regular self-audits and cross-referencing against project-specific guidelines to eliminate labeling bias and errors. Throughout the project, I collaborated with the data science team to refine the labeling taxonomy, ensuring the final dataset was optimized for model training and deployment.

2024 - 2025

Prompt Engineering and AI Training

LabelboxComputer Code ProgrammingRLHFFine Tuning

In this project, I acted as a bridge between human intent and machine execution, specializing in designing and optimizing complex prompts for Large Language Models (LLMs). My primary responsibility was to develop high-fidelity training data through Supervised Fine-Tuning (SFT), where I crafted hundreds of 'gold standard' prompt-response pairs that taught the model how to handle technical nuances, maintain a specific brand persona, and follow multi-step instructions without losing context. I played a critical role in the RLHF (Reinforcement Learning from Human Feedback) pipeline, where I evaluated and ranked model outputs based on a rigorous rubric of accuracy, safety, and helpfulness. To improve model reliability, I employed advanced prompting techniques such as Chain-of-Thought (CoT) to enhance the AI’s reasoning capabilities and Few-Shot prompting to establish pattern recognition. Additionally, I conducted 'red-teaming' sessions to identify potential hallucinations and logic gaps, i

In this project, I acted as a bridge between human intent and machine execution, specializing in designing and optimizing complex prompts for Large Language Models (LLMs). My primary responsibility was to develop high-fidelity training data through Supervised Fine-Tuning (SFT), where I crafted hundreds of 'gold standard' prompt-response pairs that taught the model how to handle technical nuances, maintain a specific brand persona, and follow multi-step instructions without losing context. I played a critical role in the RLHF (Reinforcement Learning from Human Feedback) pipeline, where I evaluated and ranked model outputs based on a rigorous rubric of accuracy, safety, and helpfulness. To improve model reliability, I employed advanced prompting techniques such as Chain-of-Thought (CoT) to enhance the AI’s reasoning capabilities and Few-Shot prompting to establish pattern recognition. Additionally, I conducted 'red-teaming' sessions to identify potential hallucinations and logic gaps, i

2023 - 2025

Video Annotation

LabelboxVideoMappingFine Tuning

In this project, I performed high-precision video annotation focused on temporal tracking and object classification for a dataset of over 200 hours of footage. My core responsibility involved using interpolation techniques to maintain consistent bounding boxes across moving sequences, ensuring that unique IDs remained accurate throughout the duration of each clip. I handled complex scenarios such as object occlusion and varying lighting conditions, consistently meeting a 99% quality benchmark. By meticulously labeling frame-by-frame transitions and defining action attributes, I provided the high-fidelity ground truth data necessary for training robust motion-prediction models.

In this project, I performed high-precision video annotation focused on temporal tracking and object classification for a dataset of over 200 hours of footage. My core responsibility involved using interpolation techniques to maintain consistent bounding boxes across moving sequences, ensuring that unique IDs remained accurate throughout the duration of each clip. I handled complex scenarios such as object occlusion and varying lighting conditions, consistently meeting a 99% quality benchmark. By meticulously labeling frame-by-frame transitions and defining action attributes, I provided the high-fidelity ground truth data necessary for training robust motion-prediction models.

2023 - 2025

Rubric-Grading

Scale AITextPrompt Response Writing SFT

I served as a human evaluator for a Reinforcement Learning from Human Feedback (RLHF) project, where I was responsible for scoring and ranking AI-generated responses based on a multi-dimensional rubric. My work involved evaluating complex text outputs for accuracy, truthfulness, and tone, ensuring that the model's responses aligned with strict safety and helpfulness guidelines. I meticulously applied a 5-point grading scale to various criteria, including logical coherence and hallucination detection, often providing written justifications for scores to guide model fine-tuning. By identifying subtle nuances in language—such as passive-aggressive tones or factual inconsistencies—I helped refine the reward model, directly contributing to a 15% improvement in the AI's response quality and policy adherence.

I served as a human evaluator for a Reinforcement Learning from Human Feedback (RLHF) project, where I was responsible for scoring and ranking AI-generated responses based on a multi-dimensional rubric. My work involved evaluating complex text outputs for accuracy, truthfulness, and tone, ensuring that the model's responses aligned with strict safety and helpfulness guidelines. I meticulously applied a 5-point grading scale to various criteria, including logical coherence and hallucination detection, often providing written justifications for scores to guide model fine-tuning. By identifying subtle nuances in language—such as passive-aggressive tones or factual inconsistencies—I helped refine the reward model, directly contributing to a 15% improvement in the AI's response quality and policy adherence.

2023 - 2024

Education

Work History