Sheshanshu Kumar

Key Skills

Software

Telus

Internal/Proprietary Tooling

SuperAnnotate

Labelbox

Top Subject Matter

No subject matter listed

Top Data Types

Computer Code Programming

Image

Text

Top Label Types

Computer Programming Coding

Data Collection

Evaluation Rating

Object Detection

Prompt Response Writing SFT

Freelancer Overview

I am a skilled AI Training Data Specialist with hands-on experience in annotating and evaluating multimodal datasets, including text, images, videos, and audio. My work has focused on Computer Programming/Coding and maintaining high data quality, consistency, and confidentiality while supporting Machine Learning model development across various domains. I’ve contributed to projects involving conversational AI, driver safety systems, and medical imaging, where I performed sentiment tagging, dialogue evaluation, context validation, and vision-based annotation tasks. With a strong foundation in AI and Machine Learning, I bring a deep understanding of how accurate labels directly impact model performance. I am proficient in conducting QA checks, reviewing complex datasets, and generating feedback to improve labeling workflows and annotation guidelines. My technical background in computer vision and NLP allows me to interpret real-world data effectively, ensuring alignment with project objectives. I am detail-oriented, adaptable, and committed to delivering high-quality annotations that help AI systems learn more efficiently and responsibly.

Labeling Experience

Pontius A vs B Evaluation

LabelboxTextQuestion AnsweringText Generation

Worked on a large-scale A vs B response evaluation project for Large Language Models using the Pontius framework. The project focused on comparing paired AI-generated responses to the same prompt and selecting the better output based on predefined quality dimensions. Key responsibilities included: • Evaluating A/B model responses for relevance, correctness, completeness, reasoning quality, tone, and safety • Applying RLHF-aligned judgment criteria to rank outputs and provide preference signals • Identifying hallucinations, factual errors, bias, policy violations, and instruction-following issues • Performing fine-grained qualitative assessment rather than isolated scoring to ensure non-decomposable evaluation • Writing concise justifications explaining why one response outperformed the other Strict quality standards were followed, including guideline adherence, consistency checks, and reviewer calibration to ensure high inter-annotator agreement.

Worked on a large-scale A vs B response evaluation project for Large Language Models using the Pontius framework. The project focused on comparing paired AI-generated responses to the same prompt and selecting the better output based on predefined quality dimensions. Key responsibilities included: • Evaluating A/B model responses for relevance, correctness, completeness, reasoning quality, tone, and safety • Applying RLHF-aligned judgment criteria to rank outputs and provide preference signals • Identifying hallucinations, factual errors, bias, policy violations, and instruction-following issues • Performing fine-grained qualitative assessment rather than isolated scoring to ensure non-decomposable evaluation • Writing concise justifications explaining why one response outperformed the other Strict quality standards were followed, including guideline adherence, consistency checks, and reviewer calibration to ensure high inter-annotator agreement.

2025

Data Trainer

SuperannotateComputer Code ProgrammingQuestion AnsweringText Generation

Designing complex, domain-specific programming prompts (Python, algorithms, data structures, ML concepts) with accurate, well-structured reference responses. Evaluating and rating AI-generated outputs for correctness, logic, efficiency, edge-case handling, and adherence to instructions. Performing quality assurance checks to identify hallucinations, logical gaps, bias, or unsafe outputs, and providing corrective feedback. Supporting supervised fine-tuning (SFT) and RLHF-style evaluation workflows by ranking responses and validating expected outputs. Ensuring strict compliance with annotation guidelines, consistency standards, and accuracy benchmarks to maintain dataset reliability at scale.

Designing complex, domain-specific programming prompts (Python, algorithms, data structures, ML concepts) with accurate, well-structured reference responses. Evaluating and rating AI-generated outputs for correctness, logic, efficiency, edge-case handling, and adherence to instructions. Performing quality assurance checks to identify hallucinations, logical gaps, bias, or unsafe outputs, and providing corrective feedback. Supporting supervised fine-tuning (SFT) and RLHF-style evaluation workflows by ranking responses and validating expected outputs. Ensuring strict compliance with annotation guidelines, consistency standards, and accuracy benchmarks to maintain dataset reliability at scale.

2025

Project Sonic

TelusAudioText GenerationText Summarization

Working on Project Sonic to annotate and evaluate audio data for training voice-based AI models. Tasks include assessing speaker intent, sentiment, and acoustic quality while classifying utterances based on context and user emotion. Ensuring accurate transcription alignment and consistent tagging criteria through regular QA checks. Contributing to improved model understanding of real-world speech variations, accents, tone, and background noise scenarios.

Working on Project Sonic to annotate and evaluate audio data for training voice-based AI models. Tasks include assessing speaker intent, sentiment, and acoustic quality while classifying utterances based on context and user emotion. Ensuring accurate transcription alignment and consistent tagging criteria through regular QA checks. Contributing to improved model understanding of real-world speech variations, accents, tone, and background noise scenarios.

2025

Project Sonic

TelusImageBounding BoxObject Detection

Contributing to Project Sonic by annotating large-scale image datasets used to train computer vision models. Responsibilities include precise bounding box labeling, object classification, and quality validation to ensure accuracy and consistency across batches. Adhered to Telus guidelines for labeling standards, confidentiality, and production workflow efficiency. Supported QA improvements by identifying edge cases and providing feedback for instruction refinement.

Contributing to Project Sonic by annotating large-scale image datasets used to train computer vision models. Responsibilities include precise bounding box labeling, object classification, and quality validation to ensure accuracy and consistency across batches. Adhered to Telus guidelines for labeling standards, confidentiality, and production workflow efficiency. Supported QA improvements by identifying edge cases and providing feedback for instruction refinement.

2025

Project Sonic

TelusTextQuestion AnsweringText Generation

Working on Project Sonic under Telus, contributing to text-based training data for improving conversational AI systems. Responsibilities include sentiment and intent classification, context accuracy verification, and rating model-generated responses for relevance, helpfulness, and safety. Performed rigorous QA checks to ensure linguistic consistency and compliance with annotation guidelines. Helping enhance NLP model performance by providing precise, high-quality human annotations across varied real-world content.

Working on Project Sonic under Telus, contributing to text-based training data for improving conversational AI systems. Responsibilities include sentiment and intent classification, context accuracy verification, and rating model-generated responses for relevance, helpfulness, and safety. Performed rigorous QA checks to ensure linguistic consistency and compliance with annotation guidelines. Helping enhance NLP model performance by providing precise, high-quality human annotations across varied real-world content.

2025

Show more

Education

Work History