Md. Maruf H.

Key Skills

Software

Label Studio

CVAT

Roboflow

Internal/Proprietary Tooling

Top Subject Matter

Technology / Computer Software

Agentic AI reasoning and LLM alignment

Education / E-Learning

Top Data Types

Image

Text

Document

Top Task Types

Classification

Computer Programming/Coding

Fine-tuning

Function Calling

RLHF

Evaluation/Rating

Prompt + Response Writing (SFT)

Data Collection

Freelancer Overview

Data Preparation & Annotation – Butterfly Project. Core strengths include Internal and Proprietary Tooling. Education includes Bachelor of Science, University of Rajshahi (2024). AI-training focus includes data types such as Image and Text and labeling workflows including Classification, Evaluation, and Rating.

Labeling Experience

AI Model Evaluation & Prompt Engineering – NEUCLEUS Project

Text

During the NEUCLEUS project, I rigorously tested and evaluated AI model outputs. I crafted intricate prompts to probe agent reasoning, rated model responses for both factual accuracy and safety, and debugged unexpected behavior to guide future training. This process closely mirrors real-world LLM evaluation and RLHF tasks. • Developed adversarial prompt scenarios for LLM/agent testing. • Provided structured ratings and qualitative feedback on model outputs. • Analyzed and reported response failures to inform system improvements. • Supported iterative enhancement of model safety and reliability.

2024 - 2024

Data Preparation & Annotation – Butterfly Project

ImageClassification

I verified and cleaned labeled image datasets for plant disease detection as part of the Butterfly project. My responsibilities included conducting error analysis on model predictions to improve annotation accuracy and quality. Additionally, I created structured prompt-response pairs for supervised LLM fine-tuning tasks. • Managed manual validation of labeled images for a rice leaf dataset. • Enhanced data quality through thorough error review and relabeling processes. • Developed SFT datasets by formulating reliable prompt-response pairs. • Directly improved downstream model performance through refined annotations.

2024 - 2024

Education

Work History