Varun Karkera

Key Skills

Software

CVAT

Labelbox

LabelImg

Label Studio

OpenCV AI Kit (OAK)

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Text

Image

Top Label Types

RLHF

Evaluation Rating

Prompt Response Writing SFT

Bounding Box

Segmentation

Object Detection

Freelancer Overview

I’ve worked hands on with AI training data for several years, mostly in NLP and large language model projects. I’ve helped create annotation guidelines, reviewed and cleaned labeled datasets, and worked closely with annotators to make sure the data was consistent and actually useful for model training. I’ve handled tasks like text classification, sentiment tagging, conversation evaluation, and improving prompt response quality for generative models. What really sets me apart is that I don’t just label data and move on, I think about how those labels will shape the model’s behavior. Since I understand the modeling side too, I’m always looking at edge cases, clarity, and overall data quality to make sure the end result genuinely improves performance.

Labeling Experience

LLM Instruction Tuning and Human Feedback Annotation

LabelboxTextRLHFEvaluation Rating

Worked on large scale instruction tuning and human feedback data generation for a conversational AI model. The project involved writing high quality prompt and response pairs, ranking multiple model outputs based on helpfulness, accuracy, and safety, and providing structured feedback for reinforcement learning from human feedback workflows. I also reviewed annotator work to ensure consistency and reduce bias across datasets. The dataset included over 200,000 prompt response pairs spanning general knowledge, reasoning, coding, and long form generation tasks. Strict quality guidelines were followed, including double review sampling, inter annotator agreement checks, and rubric based scoring to maintain consistency and reliability.

Worked on large scale instruction tuning and human feedback data generation for a conversational AI model. The project involved writing high quality prompt and response pairs, ranking multiple model outputs based on helpfulness, accuracy, and safety, and providing structured feedback for reinforcement learning from human feedback workflows. I also reviewed annotator work to ensure consistency and reduce bias across datasets. The dataset included over 200,000 prompt response pairs spanning general knowledge, reasoning, coding, and long form generation tasks. Strict quality guidelines were followed, including double review sampling, inter annotator agreement checks, and rubric based scoring to maintain consistency and reliability.

2024 - 2025

Computer Vision Dataset Annotation for Retail Shelf Analytics

LabelboxImageBounding BoxSegmentation

Annotated large scale retail shelf images to support an object detection model for product recognition and stock monitoring. Responsibilities included drawing precise bounding boxes around products, performing semantic segmentation for shelf regions, and verifying label accuracy across thousands of images. The dataset consisted of approximately 80,000 high resolution store images with more than 500 unique product SKUs. Quality control included peer review cycles, spot checks, and adherence to strict annotation guidelines to ensure consistent box placement, accurate class labeling, and minimal overlap errors.

Annotated large scale retail shelf images to support an object detection model for product recognition and stock monitoring. Responsibilities included drawing precise bounding boxes around products, performing semantic segmentation for shelf regions, and verifying label accuracy across thousands of images. The dataset consisted of approximately 80,000 high resolution store images with more than 500 unique product SKUs. Quality control included peer review cycles, spot checks, and adherence to strict annotation guidelines to ensure consistent box placement, accurate class labeling, and minimal overlap errors.

2022 - 2024

Education

Work History