Lazarus Danjuma

Key Skills

Software

Lionbridge

Telus

Label Studio

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Image

Video

Top Task Types

Text Generation

Action Recognition

Audio Recording

Object Detection

Evaluation Rating

Freelancer Overview

I am a certified AI Data Specialist with over five years of experience in data annotation, quality assurance, and large language model (LLM) training for leading global technology companies, including Nvidia, Google, Microsoft, and HubSpot. My expertise spans text, image, audio, and video annotation, as well as dataset management and quality evaluation for projects in NLP and conversational AI. I am skilled in aligning annotation standards with machine learning objectives, ensuring precise and high-quality datasets that drive model performance. With a strong background in English language and computer science, I excel at creative prompt engineering, data preprocessing, and multilingual data support, consistently delivering reliable results for complex AI initiatives.

Labeling Experience

Bandwith Audio Assessment

LionbridgeAudioText GenerationAction Recognition

The primary goal of this assessment is to evaluate the perceptual quality of audio under different "bandwidth" constraints. This data is used to improve how AI voice assistants (like Siri, Alexa, or Google Assistant) or VOIP systems perform when a user has a poor internet connection. Objective: To distinguish between "high bandwidth" (HD voice, studio quality) and "low bandwidth" (telephone quality, compressed, artifact-heavy) audio. Target: Training models to either clean this audio (noise cancellation) or understand it better despite the low quality.

The primary goal of this assessment is to evaluate the perceptual quality of audio under different "bandwidth" constraints. This data is used to improve how AI voice assistants (like Siri, Alexa, or Google Assistant) or VOIP systems perform when a user has a poor internet connection. Objective: To distinguish between "high bandwidth" (HD voice, studio quality) and "low bandwidth" (telephone quality, compressed, artifact-heavy) audio. Target: Training models to either clean this audio (noise cancellation) or understand it better despite the low quality.

2025 - 2025

Project Motion

Label StudioVideoAction RecognitionEvaluation Rating

Objective: Train AI models to understand movement, temporal context, and behavior over time (not just static objects). Target: Surveillance (anomaly detection), Autonomous Driving (predicting motion paths), and Sports Analytics (player tracking). 2. Data Labeling Tasks Object Tracking (Interpolation): Drawing a bounding box on an object in Frame 1 and Frame 10, then letting the software auto-fill the path in between. Temporal Event Tagging: Marking the precise Start (00:02) and End (00:05) timestamps of a specific action (e.g., "Car braking," "Player passing ball"). Activity Classification: Labeling the nature of the scene (e.g., "Aggressive Behavior" vs. "Normal Conversation").

Objective: Train AI models to understand movement, temporal context, and behavior over time (not just static objects). Target: Surveillance (anomaly detection), Autonomous Driving (predicting motion paths), and Sports Analytics (player tracking). 2. Data Labeling Tasks Object Tracking (Interpolation): Drawing a bounding box on an object in Frame 1 and Frame 10, then letting the software auto-fill the path in between. Temporal Event Tagging: Marking the precise Start (00:02) and End (00:05) timestamps of a specific action (e.g., "Car braking," "Player passing ball"). Activity Classification: Labeling the nature of the scene (e.g., "Aggressive Behavior" vs. "Normal Conversation").

2025

Project Chronos

TelusImageObject DetectionAction Recognition

The goal of these projects is to teach AI models how to "see" and interpret visual data. Objective: To identify, locate, and classify objects within an image with high precision. Target: Training models for Object Detection (Where is it?), Classification (What is it?), and Segmentation (What is its exact shape?). Common use cases include self-driving cars (identifying pedestrians vs. lampposts) or retail (identifying products on a shelf). 2. Data Labeling Tasks Performed Raters typically perform one or a combination of the following tasks: 2D Bounding Boxes: Drawing a tight rectangle around an object (e.g., a car). This is the most common task. Polygons: Plotting points to trace the exact outline of an irregular shape (e.g., a tree, a cloud, or a spill on the floor). This is more time-consuming but more accurate. Keypoint Annotation: Marking specific points on an object, such as the corners of a mouth for facial recognition or joints (elbows, knees) for human pose estimation.

The goal of these projects is to teach AI models how to "see" and interpret visual data. Objective: To identify, locate, and classify objects within an image with high precision. Target: Training models for Object Detection (Where is it?), Classification (What is it?), and Segmentation (What is its exact shape?). Common use cases include self-driving cars (identifying pedestrians vs. lampposts) or retail (identifying products on a shelf). 2. Data Labeling Tasks Performed Raters typically perform one or a combination of the following tasks: 2D Bounding Boxes: Drawing a tight rectangle around an object (e.g., a car). This is the most common task. Polygons: Plotting points to trace the exact outline of an irregular shape (e.g., a tree, a cloud, or a spill on the floor). This is more time-consuming but more accurate. Keypoint Annotation: Marking specific points on an object, such as the corners of a mouth for facial recognition or joints (elbows, knees) for human pose estimation.

2025

Education

Work History