Soumik T.

Key Skills

Software

AWS SageMaker

Top Subject Matter

LLM training (coding, STEM, creative) and RLHF data preparation

Multimodal dataset annotation and response evaluation

Legal Services & Contract Review

Top Data Types

Text

Image

Document

Top Task Types

RLHF

Entity (NER) Classification

Tracking

Classification

Freelancer Overview

AI Trainer & Prompt Engineer (Contractor) | Outlier AI. Brings 7+ years of professional experience across legal operations, contract review, compliance, and structured analysis. Core strengths include Internal, Proprietary Tooling, and AWS SageMaker. Education includes Bachelor of Technology, St. Thomas' College of Engineering & Technology (2019) and Higher Secondary, South Point High School (2015). AI-training focus includes data types such as Text and Image and labeling workflows including RLHF, Named Entity Recognition, and Evaluation.

Labeling Experience

AI Trainer & Prompt Engineer (Contractor) | Outlier AI

TextRLHFEntity (NER) Classification

Designed, evaluated, and refined prompts for LLM training datasets across coding, STEM, and creative tasks. Conducted RLHF-based ranking and annotation to improve model alignment and factuality. Developed golden solutions for benchmark test cases used in model fine-tuning. • Maintained quality scores above 98% across 1,000+ annotation tasks • Produced training/ranking annotations from model interactions • Worked on prompt diversity across multiple domains • Ensured dataset suitability for LLM training and evaluation

2023 - Present

Top 3% Contributor – Outlier AI (India), consistently rated 4.9/5 for annotation quality (

AWS SageMakerImageTrackingRLHF

No description provided.

2023 - 2024

AI Trainer & Data Annotator (Contractor) | Soul AI

Text

Annotated and labeled multimodal datasets including text, code, and image-text pairs for training and evaluation. Evaluated model outputs for safety, helpfulness, and instruction-following quality. Contributed to prompt diversity studies to expand coverage for edge cases. • Performed dataset labeling for multimodal instruction data • Conducted safety and quality scoring on model responses • Supported prompt variation research for robustness • Improved label coverage for harder scenarios and exceptions

2022 - 2022

Education

Work History