Sangam Rai

Key Skills

Software

CVAT

Labelbox

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Image

Text

Video

Top Label Types

Classification

Computer Programming Coding

Polygon

Segmentation

Freelancer Overview

I have experience labeling image and text datasets for AI model training, including bounding boxes, segmentation, text classification, and intent tagging. I’ve worked with structured annotation tools and followed strict QA guidelines to maintain high accuracy and consistency. I understand how high-quality data directly impacts model performance due to my technical AI background. I focus on precision, edge cases, and delivering clean, well-structured datasets ready for training and deployment.

Labeling Experience

Urban Street Scene Object Detection & Segmentation

CVATVideoPolygonSegmentation

Annotated a dataset of 5,000+ frames of urban street footage to train perception models. Tasks included drawing tight 2D bounding boxes around vehicles, pedestrians, and cyclists, as well as precise polygon segmentation for lane markings and drivable areas. Handled complex scenarios including heavy occlusion and low-light conditions. Maintained a consistent Intersection over Union (IoU) accuracy score of >95% throughout the project lifecycle.

Annotated a dataset of 5,000+ frames of urban street footage to train perception models. Tasks included drawing tight 2D bounding boxes around vehicles, pedestrians, and cyclists, as well as precise polygon segmentation for lane markings and drivable areas. Handled complex scenarios including heavy occlusion and low-light conditions. Maintained a consistent Intersection over Union (IoU) accuracy score of >95% throughout the project lifecycle.

2025 - 2025

LLM Instruction Fine-Tuning Dataset Labeling

Scale AITextClassification

Labeled 25,000+ text prompts for intent classification and response quality evaluation. Categorized user intents, flagged harmful or policy-violating content, and rated AI-generated responses for coherence, relevance, and safety. Followed structured labeling rubric to maintain inter-annotator agreement above 0.9 Cohen’s Kappa.

Labeled 25,000+ text prompts for intent classification and response quality evaluation. Categorized user intents, flagged harmful or policy-violating content, and rated AI-generated responses for coherence, relevance, and safety. Followed structured labeling rubric to maintain inter-annotator agreement above 0.9 Cohen’s Kappa.

2024 - 2024

Education

Work History