Kiran Kumari

Key Skills

Software

AWS SageMaker

Google Cloud Vertex AI

Mercor

Mindrift

CVAT

Labelbox

Top Subject Matter

No subject matter listed

Top Data Types

Computer Code Programming

Image

Text

Top Task Types

Action Recognition

Classification

Emotion Recognition

Object Detection

Prompt + Response Writing (SFT)

Freelancer Overview

I am an intermediate-level data labeling professional with hands-on experience in AI training data preparation, specializing in image classification and audio emotion recognition projects. I have worked extensively with tools such as CVAT and Labelbox to deliver high-quality, accurately annotated datasets aligned with strict annotation guidelines and quality standards. My experience includes maintaining consistent label taxonomies, performing self-reviews and quality checks, and efficiently handling large-scale image and audio datasets to support reliable model training and evaluation. I bring a strong analytical foundation from my master's degree in mathematics from IIT Delhi, which enables me to approach annotation tasks with precision, logical consistency, and attention to detail. I am committed to accuracy, reliability, and continuous learning, and I aim to contribute high-quality labeled data for robust AI and machine learning models.

Labeling Experience

Code and Prompt–Response Annotation (SFT for LLM Training):

Aws SagemakerComputer Code ProgrammingPrompt Response Writing SFT

Annotated prompt–response pairs for SFT of LLMs, focusing on machine learning and programming concepts. Evaluated code for correctness, efficiency, readability, and rubric alignment. Performed consistency checks, resolved ambiguous prompts, and refined responses for clarity and accuracy.

Annotated prompt–response pairs for SFT of LLMs, focusing on machine learning and programming concepts. Evaluated code for correctness, efficiency, readability, and rubric alignment. Performed consistency checks, resolved ambiguous prompts, and refined responses for clarity and accuracy.

2025 - 2025

Video Action Recognition and Temporal Annotation:

CVATVideoAction Recognition

Annotated video datasets for action recognition by labeling human activities across frames using temporal segments and class labels. Performed frame- and segment-level annotations to ensure accurate action start/end timing and consistency. Conducted self-reviews and quality checks while efficiently handling large-scale video datasets.

Annotated video datasets for action recognition by labeling human activities across frames using temporal segments and class labels. Performed frame- and segment-level annotations to ensure accurate action start/end timing and consistency. Conducted self-reviews and quality checks while efficiently handling large-scale video datasets.

2025 - 2025

Audio Emotion Recognition Annotation

LabelboxAudioEmotion Recognition

Annotated audio recordings for emotion recognition by accurately labeling emotional states such as happiness, sadness, anger, and neutrality based on vocal cues and annotation guidelines. Ensured consistent labeling across datasets, performed quality checks and self-reviews, and handled large volumes of audio data efficiently. The project focused on high accuracy and reliability to support training and evaluation of emotion detection models

Annotated audio recordings for emotion recognition by accurately labeling emotional states such as happiness, sadness, anger, and neutrality based on vocal cues and annotation guidelines. Ensured consistent labeling across datasets, performed quality checks and self-reviews, and handled large volumes of audio data efficiently. The project focused on high accuracy and reliability to support training and evaluation of emotion detection models

2025 - 2025

Image Classification Annotation

CVATImageClassification

Performed image-level classification using CVAT by assigning predefined class labels based on visual content and annotation guidelines. Ensured consistent label taxonomy, followed strict quality standards, and conducted self-reviews to minimize errors. Efficiently handled large-scale image datasets while maintaining high accuracy consistency to support reliable model training and evaluation

Performed image-level classification using CVAT by assigning predefined class labels based on visual content and annotation guidelines. Ensured consistent label taxonomy, followed strict quality standards, and conducted self-reviews to minimize errors. Efficiently handled large-scale image datasets while maintaining high accuracy consistency to support reliable model training and evaluation

2025 - 2025

Education

Work History