Joseph Kiarie

Key Skills

Software

Labelbox

Scale AI

CVAT

Top Subject Matter

No subject matter listed

Top Data Types

Video

Text

Image

Top Label Types

Bounding Box

Point Key Point

Segmentation

Classification

Tracking

RLHF

Freelancer Overview

I am an experienced AI Training Specialist with over three years of hands-on expertise in data annotation and labeling for machine learning projects. My background spans high-precision annotation across image, video, audio, and text datasets, supporting domains such as computer vision, speech recognition, and natural language processing. I have consistently maintained 98%+ annotation accuracy while handling large-scale datasets, including annotating and validating over 150,000 images and processing extensive audio for conversational AI systems. I am highly skilled in tools like Labelbox, CVAT, Scale AI, SuperAnnotate, and V7 Darwin, with deep experience in tasks such as object detection, segmentation, tracking, text classification, and sentiment labeling. I thrive in fast-paced, remote environments, quickly adapting to new guidelines and collaborating with distributed AI teams to deliver clean, structured training data that improves model performance and reliability.

Labeling Experience

Lead Computer Vision Annotator for Autonomous Driving (L4/L5)

LabelboxVideoBounding BoxPoint Key Point

Lead data annotation specialist for a Tier-1 autonomous vehicle training pipeline. Responsible for processing a dataset of 150,000+ video frames and images with a focus on dynamic object tracking and occlusion handling in dense urban environments. Key Deliverables: -Executed frame-by-frame 2D Bounding Box and Polygon Segmentation for vehicles, pedestrians, and cyclists. -Maintained a consistent 98% Quality Assurance (QA) score, adhering to strict pixel-perfect intersection-over-union (IoU) standards. +1 Utilized Labelbox to identify critical edge cases (e.g., severe weather reflections, sensor noise) and refined the project taxonomy to reduce model hallucinations. Acted as a QA auditor for junior annotators, ensuring guideline compliance across the workflow.

Lead data annotation specialist for a Tier-1 autonomous vehicle training pipeline. Responsible for processing a dataset of 150,000+ video frames and images with a focus on dynamic object tracking and occlusion handling in dense urban environments. Key Deliverables: -Executed frame-by-frame 2D Bounding Box and Polygon Segmentation for vehicles, pedestrians, and cyclists. -Maintained a consistent 98% Quality Assurance (QA) score, adhering to strict pixel-perfect intersection-over-union (IoU) standards. +1 Utilized Labelbox to identify critical edge cases (e.g., severe weather reflections, sensor noise) and refined the project taxonomy to reduce model hallucinations. Acted as a QA auditor for junior annotators, ensuring guideline compliance across the workflow.

2024

Image Annotator: Large-Scale Object Detection & Classification

CVATImageBounding Box

Lead annotator for a high-volume computer vision dataset focused on Object Detection for smart city analytics. Successfully annotated and validated over 150,000+ images, ensuring strict adherence to YOLO v8 formatting requirements. Key Deliverables: Created precise 2D Bounding Boxes for multi-class objects (vehicles, pedestrians, infrastructure) in dense environments. Performed Semantic Segmentation (Polygons) for irregular shapes, achieving a 99% pixel-accuracy rate during QA audits. Conducted "Ground Truth" validation to correct model predictions, significantly reducing false positives in the training set. Managed dataset attributes and metadata (occlusion, truncation levels) to improve model robustness.

Lead annotator for a high-volume computer vision dataset focused on Object Detection for smart city analytics. Successfully annotated and validated over 150,000+ images, ensuring strict adherence to YOLO v8 formatting requirements. Key Deliverables: Created precise 2D Bounding Boxes for multi-class objects (vehicles, pedestrians, infrastructure) in dense environments. Performed Semantic Segmentation (Polygons) for irregular shapes, achieving a 99% pixel-accuracy rate during QA audits. Conducted "Ground Truth" validation to correct model predictions, significantly reducing false positives in the training set. Managed dataset attributes and metadata (occlusion, truncation levels) to improve model robustness.

2023 - 2023

RLHF Specialist: Multimodal Model Evaluation & Fact-Checking

Scale AITextRLHF

Conducted high-volume Reinforcement Learning from Human Feedback (RLHF) to improve the reasoning capabilities of a large language model (LLM). Focused on evaluating model responses for factuality, safety, and coherence across complex prompt chains. Key Deliverables: Ranked and rewrote model outputs to align with "Gold Standard" human preference guidelines. - Performed Named Entity Recognition (NER) and sentiment analysis on large-scale text datasets to improve intent classification. Identified and flagged adversarial prompts (Red Teaming) to enhance model safety protocols. - Collaborated with engineering teams to refine the "Style Guide" for ambiguity reduction, reducing dataset error rates by 15%.

Conducted high-volume Reinforcement Learning from Human Feedback (RLHF) to improve the reasoning capabilities of a large language model (LLM). Focused on evaluating model responses for factuality, safety, and coherence across complex prompt chains. Key Deliverables: Ranked and rewrote model outputs to align with "Gold Standard" human preference guidelines. - Performed Named Entity Recognition (NER) and sentiment analysis on large-scale text datasets to improve intent classification. Identified and flagged adversarial prompts (Red Teaming) to enhance model safety protocols. - Collaborated with engineering teams to refine the "Style Guide" for ambiguity reduction, reducing dataset error rates by 15%.

2023 - 2023

Education

Work History