Aman Buttan

Key Skills

Software

AWS SageMaker

Top Subject Matter

No subject matter listed

Top Data Types

Computer Code Programming

Document

Text

Top Task Types

Classification

Computer Programming/Coding

Entity (NER) Classification

Relationship

Translation/Localization

Freelancer Overview

Abandoned Chat Analysis with AWS SageMaker Ground Truth: Independently delivered comprehensive data science project analyzing virtual chatbot transcripts, achieving 20% reduction in false abandoned chats through advanced data analysis and pattern recognition. Leveraged AWS SageMaker Ground Truth to create high-quality labeled training datasets from 50,000+ chat transcripts, implementing custom labeling workflows to classify chat abandonment reasons (technical issues, user frustration, resolved queries, genuine abandonment). Utilized SageMaker's active learning capabilities to reduce manual labeling effort by 40% while maintaining 95% annotation accuracy. The labeled dataset enabled training of a multi-class classification model using ensemble methods (Gradient Boosting, CatBoost) that could predict and prevent false abandonment triggers in real-time, directly improving customer experience metrics and reducing unnecessary support escalations.

Labeling Experience

Abandoned Chat Analysis with AWS SageMaker Ground Truth

Aws SagemakerTextEntity Ner ClassificationClassification

Led end-to-end data labeling project analyzing 50,000+ virtual chatbot transcripts to identify and classify chat abandonment patterns. Implemented AWS SageMaker Ground Truth workflows to create high-quality labeled training datasets for multi-class classification of abandonment reasons including: technical issues, user frustration, resolved queries, and genuine abandonment. Scope & Tasks Performed: - Set up custom labeling workflows and clear annotation guidelines - Coordinated annotation team to label chat transcripts by abandonment type - Applied quality control checks to ensure consistent and accurate labeling - Used active learning to automatically label similar data and reduce manual work - Created test datasets and monitored labeling quality throughout the project Project Size: 50,000+ chat transcripts across 4 abandonment categories

Led end-to-end data labeling project analyzing 50,000+ virtual chatbot transcripts to identify and classify chat abandonment patterns. Implemented AWS SageMaker Ground Truth workflows to create high-quality labeled training datasets for multi-class classification of abandonment reasons including: technical issues, user frustration, resolved queries, and genuine abandonment. Scope & Tasks Performed: - Set up custom labeling workflows and clear annotation guidelines - Coordinated annotation team to label chat transcripts by abandonment type - Applied quality control checks to ensure consistent and accurate labeling - Used active learning to automatically label similar data and reduce manual work - Created test datasets and monitored labeling quality throughout the project Project Size: 50,000+ chat transcripts across 4 abandonment categories

2019 - 2020

Education

Work History