Somnath Paul

Key Skills

Software

CVAT

OneForma

Top Subject Matter

No subject matter listed

Top Data Types

Image

Text

Video

Top Label Types

Bounding Box

Polygon

Segmentation

Classification

Object Detection

Entity Ner Classification

Text Summarization

RLHF

Evaluation Rating

Prompt Response Writing SFT

Action Recognition

Tracking

Freelancer Overview

I am an experienced AI Data Specialist with a strong background in data annotation, LLM evaluation, and quality analysis for AI training workflows. My expertise includes evaluating large language model outputs for factual accuracy, logical reasoning, coherence, and instruction compliance, as well as conducting comparative response analysis and structured annotation across text and image data. I am highly skilled in identifying hallucinations, bias, and edge cases, and I consistently maintain high accuracy and quality standards in remote, performance-driven environments. My technical proficiency covers AI evaluation platforms, data annotation tools, and structured reporting, and I am adept at quickly adapting to evolving guidelines and project requirements. I am passionate about contributing to advanced AI training and evaluation projects by leveraging my analytical mindset, attention to detail, and commitment to quality-focused execution.

Labeling Experience

Computer Vision Image Annotation Specialist

CVATImageBounding BoxPolygon

Worked on large-scale image annotation projects involving object detection and semantic segmentation. Labeled objects using bounding boxes and polygons, ensured pixel-level precision for segmentation tasks, and performed image classification based on predefined taxonomies. Maintained high annotation accuracy through strict adherence to project guidelines, edge-case documentation, and internal QA checks. Experienced in handling complex scenarios such as occlusion, overlapping objects, and low-visibility frames.

Worked on large-scale image annotation projects involving object detection and semantic segmentation. Labeled objects using bounding boxes and polygons, ensured pixel-level precision for segmentation tasks, and performed image classification based on predefined taxonomies. Maintained high annotation accuracy through strict adherence to project guidelines, edge-case documentation, and internal QA checks. Experienced in handling complex scenarios such as occlusion, overlapping objects, and low-visibility frames.

2024

Video Annotation & Temporal Segmentation Specialist

CVATVideoSegmentationObject Detection

Performed frame-by-frame video annotation including object tracking, action recognition, and temporal segmentation. Labeled moving objects across multiple frames while maintaining ID consistency and accurate timestamp boundaries. Handled overlapping activities and ensured precise classification of events according to project taxonomies. Maintained consistency across long video sequences and validated outputs through QA review cycles.

Performed frame-by-frame video annotation including object tracking, action recognition, and temporal segmentation. Labeled moving objects across multiple frames while maintaining ID consistency and accurate timestamp boundaries. Handled overlapping activities and ensured precise classification of events according to project taxonomies. Maintained consistency across long video sequences and validated outputs through QA review cycles.

2023

LLM & NLP Data Annotation Specialist

OneformaTextEntity Ner ClassificationText Summarization

Annotated and evaluated text datasets for training and fine-tuning large language models. Tasks included named entity recognition (NER), content categorization, response quality evaluation (RLHF), and prompt-response generation (SFT). Worked on safety, factuality, instruction-following, and tone evaluation across multilingual datasets (English, Hindi, Bengali). Ensured strict compliance with annotation guidelines and quality scoring frameworks.

Annotated and evaluated text datasets for training and fine-tuning large language models. Tasks included named entity recognition (NER), content categorization, response quality evaluation (RLHF), and prompt-response generation (SFT). Worked on safety, factuality, instruction-following, and tone evaluation across multilingual datasets (English, Hindi, Bengali). Ensured strict compliance with annotation guidelines and quality scoring frameworks.

2023

Education

Work History